# ecosystem-security

A continuous security analysis engine for software supply chains, agent skills, and connected MCP tools.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## What one organization revealed about AI agent risk

DevFeed: [What one organization revealed about AI agent risk](<https://devfeed.tech/articles/what-one-organization-revealed-about-ai-agent-risk-118937.md>)

Original publisher: [Read original article](<https://webflow.sysdig.com/blog/a-first-look-under-the-hood-of-ai-agents>)

Author: Crystal Morin

Published: 2026-11-04T00:00:00Z

Content type: news

Language: en

Sources: [Sysdig](<https://devfeed.tech/sources/sysdig-blog.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Falco](<https://devfeed.tech/topics/falco.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-credential-access](<https://devfeed.tech/tags/agent-credential-access.md>), [agentic-ai-security](<https://devfeed.tech/tags/agentic-ai-security.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agent-audit-logs](<https://devfeed.tech/tags/ai-agent-audit-logs.md>), [ai-agent-governance](<https://devfeed.tech/tags/ai-agent-governance.md>), [ai-agent-risk](<https://devfeed.tech/tags/ai-agent-risk.md>), [ai-agent-security](<https://devfeed.tech/tags/ai-agent-security.md>), [ai-agent-usage-report](<https://devfeed.tech/tags/ai-agent-usage-report.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-defense](<https://devfeed.tech/tags/ai-defense.md>), [ai-governance](<https://devfeed.tech/tags/ai-governance.md>), [ai-supply-chain-security](<https://devfeed.tech/tags/ai-supply-chain-security.md>), [audit](<https://devfeed.tech/tags/audit.md>), [citizen-developers](<https://devfeed.tech/tags/citizen-developers.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [claude-code-security](<https://devfeed.tech/tags/claude-code-security.md>), [cli](<https://devfeed.tech/tags/cli.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-permissions](<https://devfeed.tech/tags/cloud-permissions.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [code](<https://devfeed.tech/tags/code.md>), [data](<https://devfeed.tech/tags/data.md>), [data-exfiltration](<https://devfeed.tech/tags/data-exfiltration.md>), [kernel-level-visibility](<https://devfeed.tech/tags/kernel-level-visibility.md>), [mcp-server-security](<https://devfeed.tech/tags/mcp-server-security.md>), [model-context-protocol](<https://devfeed.tech/tags/model-context-protocol.md>), [risk](<https://devfeed.tech/tags/risk.md>), [runtime-security-for-ai-agents](<https://devfeed.tech/tags/runtime-security-for-ai-agents.md>), [shadow-ai](<https://devfeed.tech/tags/shadow-ai.md>), [sysdig-ai-defense](<https://devfeed.tech/tags/sysdig-ai-defense.md>), [yolo-mode](<https://devfeed.tech/tags/yolo-mode.md>), [zero-trust-for-ai-agents](<https://devfeed.tech/tags/zero-trust-for-ai-agents.md>)

### AI overview

In a month-long review using Sysdig AI Defense, one software company found that AI agent activity extended well beyond engineering: half of employees using agents were outside engineering, and 42% of those non-engineers had agents run code, builds, installs, or cloud or cluster commands. More than 60% of tool calls lacked direct human approval: 52% used Claude Code auto-mode classification and 9% had prompts disabled. Agent logs were incomplete as audit evidence: when agents read credential files, their logs omitted the action 82% of the time, while kernel monitoring captured every read; the kernel recorded seven processes per logged tool call. The organization also found that 47% of MCP servers came from third-party registries or vendors, and over half of AI spending went to duplicative capabilities. The excerpt recommends monitoring installs and CLI calls, offering non-engineers approved sandboxed build paths, setting access according to what tools and agents can reach or change, and matching model choice to tasks. It presents these as findings from one organization, not universal rates.

### Source excerpt

One company monitored AI agent activity for a month. See what Sysdig AI Defense found, from unapproved agents to unlogged credential reads.

## AI coding agents are installing MCP servers that will never be patched

DevFeed: [AI coding agents are installing MCP servers that will never be patched](<https://devfeed.tech/articles/ai-coding-agents-are-installing-mcp-servers-that-will-never-be-patched-113181.md>)

Original publisher: [Read original article](<https://www.helpnetsecurity.com/2026/11/03/mcphantom-deprecated-mcp-servers/>)

Author: Mirko Zorz

Published: 2026-11-03T05:45:53Z

Content type: news

Language: en

Sources: [Help Net Security](<https://devfeed.tech/sources/help-net-security.md>)

Topics: [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [software dark matter](<https://devfeed.tech/topics/software-dark-matter.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [trust](<https://devfeed.tech/topics/trust.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-coding](<https://devfeed.tech/tags/ai-coding.md>), [ai-coding-agent](<https://devfeed.tech/tags/ai-coding-agent.md>), [ai-coding-agents](<https://devfeed.tech/tags/ai-coding-agents.md>), [anthropic](<https://devfeed.tech/tags/anthropic.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [cursor](<https://devfeed.tech/tags/cursor.md>), [cybersecurity](<https://devfeed.tech/tags/cybersecurity.md>), [deprecated](<https://devfeed.tech/tags/deprecated.md>), [developer](<https://devfeed.tech/tags/developer.md>), [github](<https://devfeed.tech/tags/github.md>), [glow](<https://devfeed.tech/tags/glow.md>), [labs](<https://devfeed.tech/tags/labs.md>), [machines](<https://devfeed.tech/tags/machines.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-servers](<https://devfeed.tech/tags/mcp-servers.md>), [news](<https://devfeed.tech/tags/news.md>), [openai](<https://devfeed.tech/tags/openai.md>), [package](<https://devfeed.tech/tags/package.md>), [patched](<https://devfeed.tech/tags/patched.md>), [research](<https://devfeed.tech/tags/research.md>), [servers](<https://devfeed.tech/tags/servers.md>), [software-development](<https://devfeed.tech/tags/software-development.md>)

### AI overview

Glow Labs reports that coding agents installed or suggested deprecated MCP servers in the sessions it examined; in 98% of those cases, agents appeared to select package names from training data. The pattern spanned several coding tools and models. Four high-severity vulnerabilities were found in the reference servers, including a Puppeteer server flaw that can turn browser access into arbitrary command execution through prompt injection. The archived repository will not be patched, and npm deprecation warnings would not prevent npx installation. The excerpt says common controls can miss these servers because npx installs leave no dependency manifest or registered application, and one agent installed a package despite finding its deprecated status. Suggested steps include replacing deprecated servers with service-owner versions where available, approving one server per service, and including npm deprecation status in approval criteria. Harness warnings or refusal and security checks through pre-execution hooks are also proposed; coverage may vary across harnesses.

### Source excerpt

When a developer tells an AI coding agent to "set up a GitHub MCP," the agent has to pick a package. New research from Glow Labs documents sessions where it picked one out of an archived repository that will never be patched again. Source: Glow Security Reviewing real AI coding session transcripts, Glow Labs found deprecated MCP servers landing on developer machines, a pattern it calls MCPhantom. In 98% of those sessions, the agent appears ... More -> The post AI coding agents are installing MCP servers that will never be patched appeared first on Help Net Security.

## SkillDRE: Dual-Stage Red-Team Evolution of Agent Skills via Pre-Execution and Runtime Feedback

DevFeed: [SkillDRE: Dual-Stage Red-Team Evolution of Agent Skills via Pre-Execution and Runtime Feedback](<https://devfeed.tech/articles/skilldre-dual-stage-red-team-evolution-of-agent-skills-via-pre-execution-and-runtime-feedback-114006.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.32400>)

Author: Pengyu Zhu, Jingyi Yang, Yi Liu, Li Sun, Sen Su

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>)

Tags: [agent-skills](<https://devfeed.tech/tags/agent-skills.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [evolution](<https://devfeed.tech/tags/evolution.md>), [feedback](<https://devfeed.tech/tags/feedback.md>), [red-team](<https://devfeed.tech/tags/red-team.md>), [runtime](<https://devfeed.tech/tags/runtime.md>)

### AI overview

SkillDRE evolves malicious agent skill packages through a loop that combines pre-execution scanner feedback with runtime outcomes under defense. It fixes a task-conditioned malicious objective and judge rule, then revises the skill while seeking to preserve its legitimate task capability. Each runtime-guided revision returns for rescanning before another execution, addressing the risk that a scanner-passing skill may fail at runtime or that a runtime repair may trigger scanner findings. On SkillsBench across four victim models, the authors report 45.28% average attack success, 40.3% above the strongest baseline; final skills received no SkillScan findings and largely preserved benign-task performance. The abstract argues that evaluating scanning or runtime defense alone may miss attack capability. These are reported benchmark results; the supplied text does not detail methods or broader applicability.

### Source excerpt

arXiv:2609.32400v2 Announce Type: replace Abstract: Agent skills package instructions, executable code, and task-specific resources into reusable artifacts that agents can improve using execution feedback. The same mechanism also enables attackers to evolve malicious skills, making them more effective and less detectable. However, a candidate skill may pass pre-execution scanning yet fail to realize its target under runtime defenses, while a revision that repairs execution may introduce new scanner findings. We introduce SkillDRE, a fully automated framework for evolving complete malicious skill packages through a dual-stage feedback loop. Given a benign task and its associated skills, SkillDRE autonomously constructs and validates a task-conditioned malicious objective and a verifiable judge rule. It then holds both fixed while evolving the skill implementation, with preservation of legitimate task capability. SkillDRE combines scanner-guided evolution with runtime-guided refinement informed by execution outcomes observed under runtime defense. Each runtime-guided revision returns to the pre-execution stage for rescanning and further optimization before re-execution, forming a cross-stage closed loop. Evaluated on SkillsBench across four victim models, SkillDRE achieves an average attack success rate of 45.28%, exceeding the strongest baseline by 40.3%, while its final submitted skills receive no SkillScan findings and largely preserve benign-task performance. These results show that two-stage defense feedback can serve as a useful learning signal for adaptive red teaming and that evaluating either defense stage in isolation can miss the resulting attack capability. Codes is available at https://github.com/whfeLingYu/SkillDRE

## Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files

DevFeed: [Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files](<https://devfeed.tech/articles/poisoned-skills-propagating-jailbreaks-through-agent-markdown-files-107962.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.37738>)

Author: Kashyap R Manjusha, Murali Emani

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [test](<https://devfeed.tech/topics/test.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [attack-surface](<https://devfeed.tech/tags/attack-surface.md>), [configuration](<https://devfeed.tech/tags/configuration.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [jailbreaks](<https://devfeed.tech/tags/jailbreaks.md>), [markdown](<https://devfeed.tech/tags/markdown.md>), [markdown-files](<https://devfeed.tech/tags/markdown-files.md>), [skills](<https://devfeed.tech/tags/skills.md>)

### AI overview

In a LangChain question-answering agent whose system prompt was assembled from a user-provided Markdown file, a jailbreak combining persona manipulation and a GPTFUZZER-based attack raised mean attack success across seven open-weight models from 0.5% with a benign prompt to 45% on 270 forbidden questions in nine harm categories; one model reached 68.9%. Neither tested obfuscation method beat plain text: a zero-width-character payload was not more effective, while a phonetic grapheme-to-phoneme payload lowered mean success to 30%. The authors report that vulnerability varied more by model than by payload obfuscation. These results concern the tested setup and models.

### Source excerpt

arXiv:2610.37738v1 Announce Type: cross Abstract: Modern agent frameworks allow users to provide Markdown files containing system prompts, tool descriptions, and skill definitions (e.g., SKILL.md), which are incorporated into the agent's context during initialization. In this work, we investigate whether this configuration mechanism can also serve as an injection channel: can a jailbreak embedded in a Markdown skill file propagate through an agent and induce prohibited behavior? To answer this question, we construct a LangChain-based question-answering agent whose system prompt is assembled from a user-provided Markdown file. We embed a jailbreak combining persona manipulation with a GPTFUZZER-based attack and evaluate its effectiveness across seven contemporary open-weight models using 270 forbidden questions spanning nine harm categories. Under a benign system prompt, the models exhibit near-complete refusal behavior, with a mean attack success rate (ASR) of just 0.5%. In contrast, the injected skill file raises the mean ASR to 45%, reaching 68.9% for individual models. We further evaluate two payload obfuscation strategies: an invisible zero-width character-based stealth payload and a phonetic payload based on grapheme-to-phoneme transformations. Neither outperforms the plain-text injection; notably, the phonetic variant reduces mean ASR to 30%. These findings identify Markdown skill files as an underexplored attack surface for AI agents and suggest that vulnerability to configuration-based jailbreaks depends more strongly on the underlying model than on the specific payload obfuscation technique.

## CVE-2026-61542 - Gearboy advisory / Gearsystem body: Unauthenticated network exposure of MCP HTTP tools

DevFeed: [CVE-2026-61542 - Gearboy advisory / Gearsystem body: Unauthenticated network exposure of MCP HTTP tools](<https://devfeed.tech/articles/cve-2026-61542-gearboy-advisory-gearsystem-body-unauthenticated-network-exposure-of-mcp-http-tools-105497.md>)

Original publisher: [Read original article](<https://cvefeed.io/vuln/detail/CVE-2026-61542>)

Author: CVEFEED.io; github.com

Published: 2026-10-28T20:18:27Z

Content type: news

Language: en

Sources: [Latest High/Critical Vulnerabilitiy Feed](<https://devfeed.tech/sources/latest-high-critical-vulnerabilitiy-feed.md>)

Topics: [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [Cross-origin resource sharing (CORS)](<https://devfeed.tech/topics/cors.md>), [cve remediation](<https://devfeed.tech/topics/cve-remediation.md>)

Tags: [access-control](<https://devfeed.tech/tags/access-control.md>), [authorization](<https://devfeed.tech/tags/authorization.md>), [bearer](<https://devfeed.tech/tags/bearer.md>), [bind](<https://devfeed.tech/tags/bind.md>), [code-execution](<https://devfeed.tech/tags/code-execution.md>), [cve-2026-61542](<https://devfeed.tech/tags/cve-2026-61542.md>), [debugging](<https://devfeed.tech/tags/debugging.md>), [emulator](<https://devfeed.tech/tags/emulator.md>), [files](<https://devfeed.tech/tags/files.md>), [http](<https://devfeed.tech/tags/http.md>), [localhost](<https://devfeed.tech/tags/localhost.md>), [loopback](<https://devfeed.tech/tags/loopback.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-server](<https://devfeed.tech/tags/mcp-server.md>), [media](<https://devfeed.tech/tags/media.md>), [network](<https://devfeed.tech/tags/network.md>), [optional](<https://devfeed.tech/tags/optional.md>), [patch](<https://devfeed.tech/tags/patch.md>), [peer](<https://devfeed.tech/tags/peer.md>), [port](<https://devfeed.tech/tags/port.md>)

### AI overview

The optional `--mcp-http` mode exposes MCP tools to reachable network peers: the described Gearsystem transport binds to `INADDR_ANY` (normally port 7777) and dispatches POSTs to `/mcp` without authorization or `Host`/`Origin` checks. A peer can initialize a session and invoke emulator and debugging operations; the advisory bounds the reported impact to the emulator process, debugging state, and local files accepted as media, with no host code execution demonstrated. A loopback-bind control blocked network access. Gearboy 3.8.7 contains a patch.

### Source excerpt

CVE ID :CVE-2026-61542 Published : Oct. 28, 2026, 8:18 p.m. | 59 minutes ago Description :Gearboy is a Game Boy, Game Boy Color, and Super Game Boy emulator and debugger with an embedded MCP server. Gearsystem, the affected component, has an optional `--mcp-http` mode that accepts unauthenticated MCP requests from any reachable network peer despite localhost-only documentation and logging. In the described `platforms/shared/desktop/mcp/mcp_transport.h` transport, the socket binds `INADDR_ANY`, normally on port 7777, and POST requests to `/mcp` reach dispatch without `Authorization`, `Host`, or `Origin` checks. Responses and OPTIONS advertise `Access-Control-Allow-Origin: *`. A peer can initialize an MCP session and invoke exposed emulator/debugging operations; even a garbage bearer token does not block dispatch, while a loopback-bind negative control prevents network access. The reported impact is bounded to the emulator process, debugging state, and local files accepted as media, rather than demonstrated host code execution. Gearboy 3.8.7 contains a patch. Severity: 8.6 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more...

## ProbeID: Verifying MCP Tool Identity for LLM Agents under Registry Drift

DevFeed: [ProbeID: Verifying MCP Tool Identity for LLM Agents under Registry Drift](<https://devfeed.tech/articles/probeid-verifying-mcp-tool-identity-for-llm-agents-under-registry-drift-101404.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.33633>)

Author: Guanxing Chen, Sining Zhang, Haohuai He, Qiujie Lv, Yu-An Huang, Zhi-An Huang

Published: 2026-10-27T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [test](<https://devfeed.tech/topics/test.md>)

Tags: [agent-verification](<https://devfeed.tech/tags/agent-verification.md>), [agents](<https://devfeed.tech/tags/agents.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [identity](<https://devfeed.tech/tags/identity.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-tool](<https://devfeed.tech/tags/mcp-tool.md>), [reference](<https://devfeed.tech/tags/reference.md>), [registry](<https://devfeed.tech/tags/registry.md>), [tool](<https://devfeed.tech/tags/tool.md>)

### AI overview

ProbeID verifies MCP tool identity by sending schema-valid probes and comparing canonicalized responses with a trusted admission reference; it abstains when evidence is ambiguous. Under assumptions including reference integrity, probe coverage, and bounded response noise, its shared probe budget grows logarithmically with candidate count and concentration bounds limit joint misclassification risk. On cached MCP-IdentityDrift cases it achieved 0.956 hit@1, while description routers were near chance; exact comparisons tied it, MMD handled heavy value noise better, and a 32B behavioral judge was similarly accurate at higher cost. In agent-bank tests, equality rejected all reachable replacements with 64 probes and bounded distance did so with 512, while two tools with drift unreachable by bank probes always passed. In an MCP-Bench agent, verification reduced tainted tasks from 15 to 9 of 16 without changing task success. Across eleven weeks, 7 of 59 real tools changed responses despite version pins.

### Source excerpt

arXiv:2610.33633v1 Announce Type: new Abstract: Agents select tools by copyable descriptions, which cannot establish which implementation runs. ProbeID compares canonicalized responses to schema-valid probes with a trusted admission reference and abstains on ambiguity, so a squatter must copy probe-observable behavior. With the gold present, independent probes and a fixed margin, standard concentration arguments give a shared probe budget logarithmic in the candidate count and bound the joint probability of a wrong decision. On cached MCP-IdentityDrift cases, strict choice reaches \(0.956\) hit@1 while description routers fall near chance, but simpler exact comparisons tie it, MMD better tolerates heavy value noise, and a 32B behavioral judge is similarly accurate at higher cost. Random probes from agent banks and a prospective held-out study of 38 tools test theory-set thresholds. On the agent banks, equality refuses every reachable replacement from 64 probes and the bounded distance from 512, and sharper post hoc gates halve both. Two tools whose drift no bank probe reaches always pass. In an MCP-Bench agent, verification cuts tainted tasks from 15 to 9 of 16 at unchanged task success. Over eleven weeks, 7 of 59 real tools changed responses despite version pins. Verification depends on reference integrity, probe coverage, and response noise.

## Detecting Insecure Compositions of Clean Agent Skills via Capability-Flow Graph Analysis

DevFeed: [Detecting Insecure Compositions of Clean Agent Skills via Capability-Flow Graph Analysis](<https://devfeed.tech/articles/detecting-insecure-compositions-of-clean-agent-skills-via-capability-flow-graph-analysis-100279.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.32155>)

Author: Yihao Li, Chuanpu Fu, Yizhou Zhao, Miao Chen, Qi Li, Ke Xu

Published: 2026-10-26T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [Graphs](<https://devfeed.tech/topics/graphs.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [detection](<https://devfeed.tech/tags/detection.md>), [flow](<https://devfeed.tech/tags/flow.md>), [graph](<https://devfeed.tech/tags/graph.md>), [reachability-analysis](<https://devfeed.tech/tags/reachability-analysis.md>), [security](<https://devfeed.tech/tags/security.md>), [skills](<https://devfeed.tech/tags/skills.md>), [two-layer](<https://devfeed.tech/tags/two-layer.md>), [verification](<https://devfeed.tech/tags/verification.md>)

### AI overview

KnotScryer detects security risks that emerge when individually benign agent skills pass data between one another. It summarizes skills as typed capability flows, builds a Skill Capability-Flow Graph to represent cross-skill information transfer, and searches for paths carrying protected or untrusted data to insecure operations. A two-layer check replays flows for reachability and uses a task-anchored LLM audit to assess whether detected chains are semantically plausible. In an analysis of 13,598 ClawHub skills, it identified 612 insecure compositions, reported as consistently exploitable across three agent frameworks. Under GLM-5.1, the reported exploitation rate was 85.6%, compared with 16.5%-18.2% for tag-match and graph-only baselines, with an 85.3% smaller search space. The authors also report that 13.7% of detected chains used only skills labeled safe by a single-skill scanner. These findings are from the supplied abstract.

### Source excerpt

arXiv:2610.32155v1 Announce Type: new Abstract: AI agents increasingly extend their capabilities by invoking third-party skills, where each skill encapsulates instructions, scripts, and resources for agents to execute on demand. This paper reveals that deploying many benign skills together introduces compositional security risks: multiple individually clean skills can compose into safety-violation chains, because the agent passes an upstream skill's output to a downstream skill's input across invocations. The risk arises not from any single skill, but from the agent's runtime orchestration, and thus cannot be captured by existing malicious skill detection methods. To this end, we develop KnotScryer to detect insecure skill compositions by treating skills as typed capability flows. In particular, we abstract each skill into a capability summary, and construct the Skill Capability-Flow Graph (SCFG), where edges denote cross-skill information flow. Moreover, we design a search policy that tracks protected or untrusted data along handoff edges to locate paths reaching insecure operations. Finally, a two-layer verification method replays the insecure information flow for reachability analysis and applies a task-anchored LLM chain audit to confirm the semantic plausibility of the identified violation chains. We deploy KnotScryer to analyze 13,598 clean skills from ClawHub, and validate the detected chains across three agent frameworks and three LLMs. KnotScryer identifies 612 insecure compositions, which are consistently exploited across agent frameworks. Meanwhile, under GLM-5.1, it outperforms tag-match and graph-only baselines by improving the exploitation rate from 16.5%-18.2% to 85.6%, while reducing the search space by 85.3%. Furthermore, 13.7% of the detected chains consist entirely of skills labeled safe by a single-skill scanner, showing that KnotScryer effectively identifies insecure compositions that per-skill auditing cannot.

## Locking down MCP at the gateway

DevFeed: [Locking down MCP at the gateway](<https://devfeed.tech/articles/locking-down-mcp-at-the-gateway-97550.md>)

Original publisher: [Read original article](<https://developers.redhat.com/articles/2026/10/23/locking-down-mcp-at-the-gateway>)

Author: David Martin, Jason Madigan

Published: 2026-10-23T13:16:36Z

Content type: article

Language: en

Sources: [Red Hat](<https://devfeed.tech/sources/red-hat.md>), [Red Hat Developer](<https://devfeed.tech/sources/red-hat-developer.md>)

Topics: [MSP MCP](<https://devfeed.tech/topics/msp-mcp.md>), [gateway](<https://devfeed.tech/topics/gateway.md>), [Cloud Networking](<https://devfeed.tech/topics/cloud-networking.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [customer](<https://devfeed.tech/tags/customer.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [gateway](<https://devfeed.tech/tags/gateway.md>), [inventory](<https://devfeed.tech/tags/inventory.md>), [locking](<https://devfeed.tech/tags/locking.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-gateway](<https://devfeed.tech/tags/mcp-gateway.md>), [model-context-protocol](<https://devfeed.tech/tags/model-context-protocol.md>), [model-context-protocol-mcp](<https://devfeed.tech/tags/model-context-protocol-mcp.md>), [network](<https://devfeed.tech/tags/network.md>), [platform](<https://devfeed.tech/tags/platform.md>), [policy](<https://devfeed.tech/tags/policy.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [red-hat](<https://devfeed.tech/tags/red-hat.md>), [service](<https://devfeed.tech/tags/service.md>), [system](<https://devfeed.tech/tags/system.md>), [systems](<https://devfeed.tech/tags/systems.md>), [telco](<https://devfeed.tech/tags/telco.md>), [ticketing](<https://devfeed.tech/tags/ticketing.md>)

### AI overview

A shared MCP gateway should enforce separate boundaries for team configuration, network reachability, caller identity, tool permissions, and request volume. Red Hat OpenShift can isolate teams with separate gateway instances or per-team listeners on one gateway, each with its own MCPGatewayExtension and server registrations. NetworkPolicy filters workload communication, not users or tools; outbound access to MCP servers is allowed by default, so a separate policy can restrict which gateway pods reach a backend. RHCL AuthPolicy validates OIDC/JWT credentials, but authentication alone grants access to all tools behind the gateway. Authorization can check the router-provided server and tool names against identity roles, rejecting unauthorized calls with 403 before backend execution; listener policies authenticate independently. RateLimitPolicy caps traffic at gateway level, with HTTPRoute policies able to replace the default for a backend. These controls do not replace content checks, credential isolation, or provenance checks.

### Source excerpt

Model Context Protocol (MCP) lets an agent use tools across your enterprise platform. In a telco, that can include your network inventory, service assurance, ticketing, and customer systems. A shared MCP gateway makes those tools available through one endpoint. That endpoint still needs a policy boundary. A caller that has authenticated to the gateway should not automatically be able to read every system or invoke every operation. The post Locking down MCP at the gateway appeared first on Red Hat Developer.

## Best Tools for AI Agent Safety in Production (2026)

DevFeed: [Best Tools for AI Agent Safety in Production (2026)](<https://devfeed.tech/articles/best-tools-for-ai-agent-safety-in-production-2026-87102.md>)

Original publisher: [Read original article](<https://www.lyzr.ai/blog/best-tools-for-ai-agent-safety-in-production/>)

Author: Lyzr Team

Published: 2026-10-14T19:49:19Z

Content type: comparison

Language: en

Sources: [Lyzr.ai](<https://devfeed.tech/sources/lyzr-ai.md>)

Topics: [enterprise data protection](<https://devfeed.tech/topics/enterprise-data-protection.md>), [agent observability](<https://devfeed.tech/topics/agent-observability.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [Instrumentation](<https://devfeed.tech/topics/instrumentation.md>), [Memory Safety](<https://devfeed.tech/topics/memory-safety.md>), [rego](<https://devfeed.tech/topics/rego.md>), [Microsoft Agent Framework](<https://devfeed.tech/topics/microsoft-agent-framework.md>)

Tags: [agent-observability](<https://devfeed.tech/tags/agent-observability.md>), [agent-security](<https://devfeed.tech/tags/agent-security.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agent-observability](<https://devfeed.tech/tags/ai-agent-observability.md>), [ai-agent-safety](<https://devfeed.tech/tags/ai-agent-safety.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-security](<https://devfeed.tech/tags/ai-security.md>), [apache](<https://devfeed.tech/tags/apache.md>), [apache-2-0](<https://devfeed.tech/tags/apache-2-0.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [bluefield-4](<https://devfeed.tech/tags/bluefield-4.md>), [bluefield-4-dpus](<https://devfeed.tech/tags/bluefield-4-dpus.md>)

### AI overview

The article compares tools for keeping AI agents within intended safety, security, access, and operational boundaries in production. It describes five control layers--input security, tool and action controls, testing, runtime monitoring, and enterprise governance--and reviews products including OpenController, Prisma AIRS, Model Armor, OpenShell, and the Microsoft Agent Governance Toolkit. It emphasizes combining safeguards according to the risks and architecture involved.

### Source excerpt

TL;DR An agent in production receives a hidden instruction inside a retrieved document, calls a connected business tool and tries to send sensitive data somewhere...

## From zero to agent builder: Running kagent on your laptop

DevFeed: [From zero to agent builder: Running kagent on your laptop](<https://devfeed.tech/articles/from-zero-to-agent-builder-running-kagent-on-your-laptop-85205.md>)

Original publisher: [Read original article](<https://www.cncf.io/blog/2026/10/12/from-zero-to-agent-builder-running-kagent-on-your-laptop/>)

Author: Antonio Berben(Solo.io) and Felipe Vicens(Telefónica)

Published: 2026-10-12T11:30:00Z

Content type: tutorial

Language: en

Sources: [Cloud Native Computing Foundation](<https://devfeed.tech/sources/cloud-native-computing-foundation.md>)

Topics: [grab agent platform](<https://devfeed.tech/topics/grab-agent-platform.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>), [Ollama](<https://devfeed.tech/topics/ollama.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>)

Tags: [agent-builder](<https://devfeed.tech/tags/agent-builder.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [blog](<https://devfeed.tech/tags/blog.md>), [code-agent](<https://devfeed.tech/tags/code-agent.md>), [kagent](<https://devfeed.tech/tags/kagent.md>), [tools](<https://devfeed.tech/tags/tools.md>)

### AI overview

This tutorial shows how to run kagent on a local Kubernetes cluster and configure an AI agent to create other agents from plain-language requests. It describes using Ollama, limiting available tools and permissions, and requiring human approval before applying a generated manifest. It also discusses governance concerns, including identity, cost tracking, and authorization for agents.

### Source excerpt

If you have seen a demo of a no-code agent builder, you know the moment. Someone types one sentence, waits a minute, and there it is: a new agent, running, with tools. It is impressive. For...

## How to Manage Agents and Skills in a Software Factory

DevFeed: [How to Manage Agents and Skills in a Software Factory](<https://devfeed.tech/articles/how-to-manage-agents-and-skills-in-a-software-factory-85006.md>)

Original publisher: [Read original article](<https://www.port.io/blog/manage-ai-agents-skills-software-factory>)

Author: Pavan Belagatti

Published: 2026-10-11T13:58:11Z

Content type: tutorial

Language: en

Sources: [Port](<https://devfeed.tech/sources/developer-experience-platform-engineering-blog-port.md>)

Topics: [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [internal developer platform](<https://devfeed.tech/topics/internal-developer-platform.md>), [Platform Engineering](<https://devfeed.tech/topics/platform-engineering.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [governance](<https://devfeed.tech/tags/governance.md>), [permissions](<https://devfeed.tech/tags/permissions.md>), [skills](<https://devfeed.tech/tags/skills.md>), [workflows](<https://devfeed.tech/tags/workflows.md>)

### AI overview

The article explains how platform teams can manage AI agents and reusable skills in a software factory. It recommends discovering assets from source repositories, assigning owners, limiting agent permissions, versioning and certifying skills, gating agents on evaluations before production, and recording runs for audit and cost tracking.

### Source excerpt

Learn how to manage AI agents and skills in a software factory with governance, permissions, reusable workflows, and clear ownership.

## What Your Search Tool Sends Your Coding Agent

DevFeed: [What Your Search Tool Sends Your Coding Agent](<https://devfeed.tech/articles/what-your-search-tool-sends-your-coding-agent-85776.md>)

Original publisher: [Read original article](<https://upstash.com/blog/what-search-hands-your-agent>)

Author: Ramazan Onur Acar

Published: 2026-10-11T00:00:00Z

Content type: article

Language: en

Sources: [Upstash](<https://devfeed.tech/sources/upstash.md>)

Topics: [devrig](<https://devfeed.tech/topics/devrig.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [DuckDuckGo](<https://devfeed.tech/topics/duckduckgo.md>), [Algolia](<https://devfeed.tech/topics/algolia.md>)

Tags: [account-takeover](<https://devfeed.tech/tags/account-takeover.md>), [ai](<https://devfeed.tech/tags/ai.md>), [api](<https://devfeed.tech/tags/api.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [backdoor](<https://devfeed.tech/tags/backdoor.md>), [brave](<https://devfeed.tech/tags/brave.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [code](<https://devfeed.tech/tags/code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [context7](<https://devfeed.tech/tags/context7.md>), [cursor](<https://devfeed.tech/tags/cursor.md>), [fake](<https://devfeed.tech/tags/fake.md>), [firecrawl](<https://devfeed.tech/tags/firecrawl.md>), [security](<https://devfeed.tech/tags/security.md>)

### AI overview

The article compares how six search tools return potentially dangerous documentation to coding agents, then tests whether language models repeat those instructions. Search tools that return full page bodies often included risky passages, while Context7 returned fewer; when harmful passages reached models through web search, they were passed along in 35% of cases. The authors caution that the study used nine known pages and does not establish rates across the web or shipped assistants.

### Source excerpt

Most dangerous docs aren't attacks. They're honest pages telling your agent to fetch something unpinned and run it. We measured how often six search tools pass them on.

## Top 6 developer security tools for enterprise teams in 2026

DevFeed: [Top 6 developer security tools for enterprise teams in 2026](<https://devfeed.tech/articles/top-6-developer-security-tools-for-enterprise-teams-in-2026-84419.md>)

Original publisher: [Read original article](<https://www.aikido.dev/blog/top-developer-security-platforms-enterprise>)

Author: Nicholas Thomson

Published: 2026-10-09T14:00:00Z

Content type: comparison

Language: en

Sources: [Aikido Security's Blog](<https://devfeed.tech/sources/aikido-security-s-blog.md>)

Topics: [GitHub](<https://devfeed.tech/topics/github.md>), [software composition analysis](<https://devfeed.tech/topics/software-composition-analysis.md>), [snyk](<https://devfeed.tech/topics/snyk.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [ai-pentesting](<https://devfeed.tech/tags/ai-pentesting.md>), [aikido-security](<https://devfeed.tech/tags/aikido-security.md>), [application-security](<https://devfeed.tech/tags/application-security.md>), [checkmarx](<https://devfeed.tech/tags/checkmarx.md>), [developer-security-platform](<https://devfeed.tech/tags/developer-security-platform.md>), [developer-security-tools](<https://devfeed.tech/tags/developer-security-tools.md>), [devsec-tools-comparisons](<https://devfeed.tech/tags/devsec-tools-comparisons.md>), [end-of-life-dependency-patching](<https://devfeed.tech/tags/end-of-life-dependency-patching.md>), [endor-labs](<https://devfeed.tech/tags/endor-labs.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [enterprise-appsec](<https://devfeed.tech/tags/enterprise-appsec.md>), [fedramp-moderate](<https://devfeed.tech/tags/fedramp-moderate.md>), [github-advanced-security](<https://devfeed.tech/tags/github-advanced-security.md>), [iso-27001-compliance](<https://devfeed.tech/tags/iso-27001-compliance.md>), [malicious-package-detection](<https://devfeed.tech/tags/malicious-package-detection.md>), [rbac](<https://devfeed.tech/tags/rbac.md>), [reachability-analysis](<https://devfeed.tech/tags/reachability-analysis.md>), [sast](<https://devfeed.tech/tags/sast.md>), [sca](<https://devfeed.tech/tags/sca.md>), [secrets-detection](<https://devfeed.tech/tags/secrets-detection.md>), [security](<https://devfeed.tech/tags/security.md>), [snyk](<https://devfeed.tech/tags/snyk.md>), [soc-2-compliance](<https://devfeed.tech/tags/soc-2-compliance.md>), [sonarqube](<https://devfeed.tech/tags/sonarqube.md>), [sso](<https://devfeed.tech/tags/sso.md>)

### AI overview

This comparison evaluates six developer security platforms for enterprise teams, focusing on governance, deployment, signal quality, and remediation. It outlines tool capabilities and tradeoffs, then offers criteria for shortlisting and evaluating platforms.

### Source excerpt

We compare the top developer security tools for enterprise teams on signal quality, governance, and remediation: Aikido Security, Snyk, GitHub Advanced Security, SonarQube, Checkmarx, and Endor Labs Category: DevSec Tools & Comparisons

## JFrog Agent Skills Registry: Features, Use Cases & Alternatives

DevFeed: [JFrog Agent Skills Registry: Features, Use Cases & Alternatives](<https://devfeed.tech/articles/jfrog-agent-skills-registry-features-use-cases-alternatives-64606.md>)

Original publisher: [Read original article](<https://www.port.io/blog/jfrog-agent-skills-registry>)

Author: John Crowley

Published: 2026-10-04T09:51:43Z

Content type: comparison

Language: en

Sources: [Port](<https://devfeed.tech/sources/developer-experience-platform-engineering-blog-port.md>)

Topics: [Jfrog](<https://devfeed.tech/topics/jfrog.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [NemoClaw](<https://devfeed.tech/topics/nemoclaw.md>), [AI for customer service automation](<https://devfeed.tech/topics/ai-for-customer-service-automation.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-skills](<https://devfeed.tech/tags/agent-skills.md>), [agentic-sdlc](<https://devfeed.tech/tags/agentic-sdlc.md>), [alternatives](<https://devfeed.tech/tags/alternatives.md>), [assets](<https://devfeed.tech/tags/assets.md>), [attack-surface](<https://devfeed.tech/tags/attack-surface.md>), [cause](<https://devfeed.tech/tags/cause.md>), [connect](<https://devfeed.tech/tags/connect.md>), [dependencies](<https://devfeed.tech/tags/dependencies.md>), [dependency](<https://devfeed.tech/tags/dependency.md>), [developer](<https://devfeed.tech/tags/developer.md>), [developer-portal](<https://devfeed.tech/tags/developer-portal.md>), [developers](<https://devfeed.tech/tags/developers.md>), [development](<https://devfeed.tech/tags/development.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [forensic](<https://devfeed.tech/tags/forensic.md>), [jfrog](<https://devfeed.tech/tags/jfrog.md>), [registry](<https://devfeed.tech/tags/registry.md>), [skills](<https://devfeed.tech/tags/skills.md>)

### AI overview

The article compares JFrog's Agent Skills Registry with an Agentic SDLC Platform. JFrog stores, scans, signs, versions, and approves agent skills and related assets, while a platform such as Port addresses ownership, certification, permissions, and promotion toward production. It argues that teams may use both, with the registry providing vetted assets and the platform governing how teams use them.

### Source excerpt

What the JFrog Agent Skills Registry does, the problems it solves, where it fits in an Agentic SDLC stack, and its alternatives.

## How Chainguard Factory hardens and verifies AI agent skills

DevFeed: [How Chainguard Factory hardens and verifies AI agent skills](<https://devfeed.tech/articles/how-do-you-harden-an-agent-skill-the-simple-file-type-with-serious-complexities-63428.md>)

Original publisher: [Read original article](<https://www.chainguard.dev/unchained/how-do-you-harden-an-agent-skill-the-simple-file-type-with-serious-complexities>)

Published: 2026-10-01T00:00:00Z

Content type: article

Language: en

Sources: [Chainguard: Unchained](<https://devfeed.tech/sources/chainguard-unchained.md>)

Topics: [hardened ai agent skills](<https://devfeed.tech/topics/hardened-ai-agent-skills.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>)

Tags: [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [chainguard](<https://devfeed.tech/tags/chainguard.md>), [risks](<https://devfeed.tech/tags/risks.md>), [skills](<https://devfeed.tech/tags/skills.md>)

### AI overview

Chainguard describes how its Factory processes AI agent skills for security. The pipeline scans skill files, removes or remediates risks, rescans them to verify fixes and functionality, then signs and publishes approved skills. The company also monitors agent behavior and actions to detect threats that static scans may miss.

### Source excerpt

Learn how Chainguard Factory hardens AI agent skills by detecting malicious behavior, fixing risks, and verifying functionality before release.

## Laravel AI SDK and Laravel MCP Security Fixes: Update Now

DevFeed: [Laravel AI SDK and Laravel MCP Security Fixes: Update Now](<https://devfeed.tech/articles/laravel-ai-sdk-and-laravel-mcp-security-fixes-update-now-62903.md>)

Original publisher: [Read original article](<https://laravel-news.com/laravel-ai-mcp-security-advisories>)

Author: Eric L. Barnes

Published: 2026-09-30T17:11:12Z

Content type: news

Language: en

Sources: [Laravel](<https://devfeed.tech/sources/laravel.md>)

Topics: [Security](<https://devfeed.tech/topics/security.md>), [Laravel](<https://devfeed.tech/topics/laravel.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>)

Tags: [news](<https://devfeed.tech/tags/news.md>), [oauth](<https://devfeed.tech/tags/oauth.md>), [security](<https://devfeed.tech/tags/security.md>), [ssrf](<https://devfeed.tech/tags/ssrf.md>), [update](<https://devfeed.tech/tags/update.md>)

### AI overview

Security advisories cover an SSRF vulnerability in the Laravel AI SDK's Vercel and AG-UI adapters and an OAuth redirect validation issue in Laravel MCP. The fixes are available in Laravel AI SDK 1.0.1 and Laravel MCP 0.9.6 and 1.0.1. The article notes affected configurations and mitigations for apps that cannot upgrade immediately.

### Source excerpt

Security advisories for the Laravel AI SDK and Laravel MCP fix an SSRF bug and an OAuth redirect issue. Here are the affected versions and how to update. The post Laravel AI SDK and Laravel MCP Security Fixes: Update Now appeared first on Laravel News. Join the Laravel Newsletter to get Laravel articles like this directly in your inbox.

## Evo ADS Govern Agent Behavior Goes GA: Bringing MCP Usage Under Control

DevFeed: [Evo ADS Govern Agent Behavior Goes GA: Bringing MCP Usage Under Control](<https://devfeed.tech/articles/evo-ads-govern-agent-behavior-goes-ga-bringing-mcp-usage-under-control-62670.md>)

Original publisher: [Read original article](<https://snyk.io/blog/evo-ads-govern-agent-behavior-ga/>)

Author: Daniel Berman

Published: 2026-09-30T04:00:00Z

Content type: news

Language: en

Sources: [Snyk](<https://devfeed.tech/sources/blog-rss-feed-snyk.md>)

Topics: [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [cursor](<https://devfeed.tech/topics/cursor.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [GitHub Copilot](<https://devfeed.tech/topics/github-copilot.md>), [Postman Agent Mode](<https://devfeed.tech/topics/postman-agent-mode.md>), [ReAct agent Python](<https://devfeed.tech/topics/react-agent-python.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-coding-agents](<https://devfeed.tech/tags/ai-coding-agents.md>), [application-security](<https://devfeed.tech/tags/application-security.md>), [blog](<https://devfeed.tech/tags/blog.md>), [codex](<https://devfeed.tech/tags/codex.md>), [developer](<https://devfeed.tech/tags/developer.md>), [devops](<https://devfeed.tech/tags/devops.md>), [devsecops](<https://devfeed.tech/tags/devsecops.md>), [governance](<https://devfeed.tech/tags/governance.md>), [interest](<https://devfeed.tech/tags/interest.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [mcp-server](<https://devfeed.tech/tags/mcp-server.md>), [security](<https://devfeed.tech/tags/security.md>), [snyk-platform](<https://devfeed.tech/tags/snyk-platform.md>), [supply-chain-security](<https://devfeed.tech/tags/supply-chain-security.md>)

### AI overview

Evo ADS Govern Agent Behavior is generally available, beginning with MCP Governance. The capability lets security and platform teams discover MCP servers, define which are allowed, and log or block their use across Claude Code, Cursor, Codex, and GitHub Copilot. MCP connects AI coding agents to external tools and data, but does not itself determine which connections should be trusted or permitted.

### Source excerpt

Evo ADS Govern Agent Behavior is now generally available, starting with MCP Governance. Discover, approve, monitor, log, and block MCP server usage across leading AI coding agents.

## AI Agent Safety Standard Defines Independent Testing Measures

DevFeed: [AI Agent Safety Standard Defines Independent Testing Measures](<https://devfeed.tech/articles/your-ai-agent-is-safe-says-who-84761.md>)

Original publisher: [Read original article](<https://www.gendigital.com/blog/insights/leadership-perspectives/ai-safety-testing>)

Author: Jakub Křoustek

Published: 2026-09-29T14:08:51Z

Content type: article

Language: en

Sources: [Gen](<https://devfeed.tech/sources/gen.md>)

Topics: [ai-governance](<https://devfeed.tech/topics/ai-governance.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>)

Tags: [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-security](<https://devfeed.tech/tags/ai-security.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [detection](<https://devfeed.tech/tags/detection.md>), [leadership-perspectives](<https://devfeed.tech/tags/leadership-perspectives.md>)

### AI overview

The article describes an industry standard adopted through AMTSO for independently evaluating AI agents and the products that protect them. It says the methodology measures the agent's own safety baseline, distinguishes blocking attacks from relying on the model to refuse them, tracks detection and prevention separately, and counts disruption of legitimate tasks as a failure.

### Source excerpt

Gen is leading the industry to develop the first standardized benchmarks for AI protection

## Postman adds security controls for AI agents, APIs, and MCP servers

DevFeed: [Postman adds security controls for AI agents, APIs, and MCP servers](<https://devfeed.tech/articles/postman-adds-security-controls-for-ai-agents-apis-and-mcp-servers-61760.md>)

Original publisher: [Read original article](<https://www.helpnetsecurity.com/2026/09/29/postman-fabric-gateway/>)

Author: Industry News

Published: 2026-09-29T13:40:47Z

Content type: news

Language: en

Sources: [Help Net Security](<https://devfeed.tech/sources/help-net-security.md>)

Topics: [AI Integration Strategies](<https://devfeed.tech/topics/ai-integration-strategies.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [OpenAPI Specification](<https://devfeed.tech/topics/openapi.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [has](<https://devfeed.tech/tags/has.md>), [industry-news](<https://devfeed.tech/tags/industry-news.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [postman](<https://devfeed.tech/tags/postman.md>), [security](<https://devfeed.tech/tags/security.md>)

### AI overview

Postman announced the general availability of Fabric Gateway, a protocol-agnostic control plane for governing how AI agents, LLMs, and MCP servers discover and interact with APIs, tools, and other agents. The gateway is deployed in an organization's own cloud and provides a workflow for discovery, identity, policy, and governance. Its described capabilities include a unified registry, least-privilege policies across protocols, and centralized oversight of agent actions and identities.

### Source excerpt

Postman has announced the general availability of Fabric Gateway, a protocol-agnostic control plane for governing how AI agents, LLMs, and MCP servers discover and interact with APIs, tools, and other agents. With Fabric Gateway, organizations can securely connect AI agents to their APIs while centrally controlling what agents can discover, access, and do without introducing fragmented tools or one-off integrations. As enterprises rapidly deploy AI agents, they are discovering that existing infrastructure was designed for ... More -> The post Postman adds security controls for AI agents, APIs, and MCP servers appeared first on Help Net Security.

## CoPhish and the Agentic Consent Problem: OAuth Phishing for the Agent-Builder Era

DevFeed: [CoPhish and the Agentic Consent Problem: OAuth Phishing for the Agent-Builder Era](<https://devfeed.tech/articles/cophish-and-the-agentic-consent-problem-oauth-phishing-for-the-agent-builder-era-61990.md>)

Original publisher: [Read original article](<https://workos.com/blog/cophish-and-the-agentic-consent-problem-oauth-phishing-for-the-agent-builder-era>)

Author: WorkOS

Published: 2026-09-29T00:00:00Z

Content type: article

Language: en

Sources: [WorkOS](<https://devfeed.tech/sources/workos-blog.md>)

Topics: [OAuth](<https://devfeed.tech/topics/oauth.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [computer-use](<https://devfeed.tech/topics/computer-use.md>)

Tags: [agent-auth](<https://devfeed.tech/tags/agent-auth.md>), [attack](<https://devfeed.tech/tags/attack.md>), [consent](<https://devfeed.tech/tags/consent.md>), [exfiltration](<https://devfeed.tech/tags/exfiltration.md>), [oauth](<https://devfeed.tech/tags/oauth.md>)

### AI overview

CoPhish used Microsoft Copilot Studio as a trusted delivery platform for an OAuth consent phishing attack. An agent could direct users to a malicious consent request and then forward the resulting token to an attacker-controlled server. The article describes exposure under default consent policies and discusses implications for builders of MCP servers and agent platforms.

### Source excerpt

CoPhish showed that an agent-building platform can itself become an OAuth phishing vector. Here's what that means for anyone building MCP servers or agent platforms, how to configure against it, and what Entra Agent ID and WorkOS Agent Auth reveal about where the industry is headed.

## Bifrost vulnerabilities expose gaps in MCP client authentication

DevFeed: [Bifrost vulnerabilities expose gaps in MCP client authentication](<https://devfeed.tech/articles/bifrost-s-mcp-flaw-shows-the-protocol-still-has-no-client-authentication-story-61991.md>)

Original publisher: [Read original article](<https://workos.com/blog/mcp-client-authentication-gap>)

Author: WorkOS

Published: 2026-09-29T00:00:00Z

Content type: article

Language: en

Sources: [WorkOS](<https://devfeed.tech/sources/workos-blog.md>)

Topics: [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [WebSocket](<https://devfeed.tech/topics/websocket.md>), [Authentication](<https://devfeed.tech/topics/authentication.md>), [HashiCorp Vault](<https://devfeed.tech/topics/hashicorp-vault.md>)

Tags: [authentication](<https://devfeed.tech/tags/authentication.md>), [bifrost](<https://devfeed.tech/tags/bifrost.md>), [cve](<https://devfeed.tech/tags/cve.md>), [docker](<https://devfeed.tech/tags/docker.md>), [http](<https://devfeed.tech/tags/http.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [vulnerabilities](<https://devfeed.tech/tags/vulnerabilities.md>)

### AI overview

Two critical unauthenticated remote code execution vulnerabilities in the Bifrost AI gateway shared a root cause: network exposure combined with dashboard authentication left disabled by default. The article argues that MCP's lack of guidance for gateway management surfaces leaves their security to application teams, and notes that Bifrost 2.1.0 requires a setup token before the first admin account can be claimed.

### Source excerpt

Two critical vulnerabilities, one root cause, and the piece of MCP that is supposed to close this gap.

## New in LangSmith Engine: red teaming and automated testing

DevFeed: [New in LangSmith Engine: red teaming and automated testing](<https://devfeed.tech/articles/new-in-langsmith-engine-red-teaming-and-automated-testing-78327.md>)

Original publisher: [Read original article](<https://www.langchain.com/blog/langsmith-engine-v2-redteam>)

Author: LangChain Accounts

Published: 2026-09-24T17:22:02Z

Content type: release

Language: en

Sources: [LangChain Blog](<https://devfeed.tech/sources/langchain-blog.md>)

Topics: [axe DevTools](<https://devfeed.tech/topics/axe-devtools.md>), [Test automation](<https://devfeed.tech/topics/test-automation.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [Monitoring](<https://devfeed.tech/topics/monitoring.md>)

Tags: [agent-testing](<https://devfeed.tech/tags/agent-testing.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [langsmith](<https://devfeed.tech/tags/langsmith.md>), [red-teaming](<https://devfeed.tech/tags/red-teaming.md>), [testing](<https://devfeed.tech/tags/testing.md>)

### AI overview

LangSmith Engine v2 adds proactive red teaming to identify agent bugs and prompt issues, detects inefficient agent paths and trends in errors, latency, and cost, and tests proposed fixes before human review. Red teaming and fix validation are in private beta for existing LangSmith Deployment users; Engine v2 is available in SaaS for Plus and Enterprise plans.

### Source excerpt

LangSmith Engine now includes Red Teaming to proactively detect agent issues and automated agent testing. Learn more about the Engine v2 release.

## A note on the Hugging Face agent incident

DevFeed: [A note on the Hugging Face agent incident](<https://devfeed.tech/articles/a-note-on-the-hugging-face-agent-incident-79823.md>)

Original publisher: [Read original article](<https://modal.com/blog/a-note-on-the-hugging-face-agent-incident>)

Author: Modal

Published: 2026-07-29T00:00:00Z

Content type: opinion

Language: en

Sources: [Modal](<https://devfeed.tech/sources/modal.md>)

Topics: [openai-hugging-face-incident](<https://devfeed.tech/topics/openai-hugging-face-incident.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [incident management](<https://devfeed.tech/topics/incident-management.md>)

Tags: [container](<https://devfeed.tech/tags/container.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [note](<https://devfeed.tech/tags/note.md>), [sandbox](<https://devfeed.tech/tags/sandbox.md>), [security](<https://devfeed.tech/tags/security.md>)

### AI overview

Modal says its platform and sandbox isolation were not compromised in the Hugging Face agent intrusion. The code execution occurred in a customer's publicly accessible application, inside that customer's container; Modal says no other customer workloads were affected. Modal recommends requiring authentication for public endpoints, limiting IP access and outbound networking, and treating user-submitted code or input as untrusted.

### Source excerpt

Hugging Face published a technical timeline of a recent agent intrusion. Modal's platform and isolation were not compromised in this incident.

## Packagist is now protected by Aikido Intel and other updates to the PHP registry

DevFeed: [Packagist is now protected by Aikido Intel and other updates to the PHP registry](<https://devfeed.tech/articles/packagist-is-now-protected-by-aikido-intel-and-other-updates-to-the-php-registry-51613.md>)

Original publisher: [Read original article](<https://www.aikido.dev/blog/composer-protected-aikido-packagist>)

Author: Dania Durnas

Published: 2026-06-26T07:52:00Z

Content type: news

Language: en

Sources: [Aikido Security's Blog](<https://devfeed.tech/sources/aikido-security-s-blog.md>)

Topics: [Composer](<https://devfeed.tech/topics/composer.md>), [Package Management](<https://devfeed.tech/topics/package-management.md>), [software supply-chain attack](<https://devfeed.tech/topics/software-supply-chain-attack.md>), [Malware](<https://devfeed.tech/topics/malware.md>), [PHP](<https://devfeed.tech/topics/php.md>), [ecosystem-security](<https://devfeed.tech/topics/ecosystem-security.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [composer](<https://devfeed.tech/tags/composer.md>), [malware](<https://devfeed.tech/tags/malware.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [package-registries](<https://devfeed.tech/tags/package-registries.md>), [php](<https://devfeed.tech/tags/php.md>), [registry](<https://devfeed.tech/tags/registry.md>), [security](<https://devfeed.tech/tags/security.md>), [supply-chain](<https://devfeed.tech/tags/supply-chain.md>), [supply-chain-attacks](<https://devfeed.tech/tags/supply-chain-attacks.md>)

### AI overview

Aikido's malware feed now blocks known-malicious package versions in Composer 2.10 and later by default. The article explains how Packagist's dependency policy framework removes flagged versions from dependency resolution, including versions identified after a lockfile was created, while allowing manual overrides.

### Source excerpt

Aikido's malware feed now blocks bad package versions in Composer by default. A look at how Packagist is closing whole classes of supply chain attacks. Category: Product & Company Updates

[Next page](<https://devfeed.tech/topics/ecosystem-security.md?cursor=WyIyMDI2LTA2LTI2VDA3OjUyOjAwKzAwOjAwIiwgImIyNmVjYmQwLWM4ZTEtNDM3ZS05NjMxLTJjMzNiZWU2ZWFkMiJd>)