# AI agent at the wheel: How an attacker used LLMs to move from a CVE to an internal database in 4 pivots

DevFeed: [AI agent at the wheel: How an attacker used LLMs to move from a CVE to an internal database in 4 pivots](<https://devfeed.tech/articles/ai-agent-at-the-wheel-how-an-attacker-used-llms-to-move-from-a-cve-to-an-internal-database-in-4-pivots-53195.md>)

Original publisher: [Read original article](<https://webflow.sysdig.com/blog/ai-agent-at-the-wheel-how-an-attacker-used-llms-to-move-from-a-cve-to-an-internal-database-in-4-pivots>)

Author: Michael Clark

Published: 2026-05-26T00:00:00Z

Content type: article

Language: en

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

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Threat Research](<https://devfeed.tech/topics/threat-research.md>), [CVE-2026-39987](<https://devfeed.tech/topics/cve-2026-39987.md>), [Vulnerabilities](<https://devfeed.tech/topics/vulnerabilities.md>), [Cloudflare Workers](<https://devfeed.tech/topics/cloudflare-workers.md>), [Amazon Web Services](<https://devfeed.tech/topics/aws.md>), [PostgreSQL](<https://devfeed.tech/topics/postgresql.md>), [OpenSSH](<https://devfeed.tech/topics/openssh.md>)

Tags: [adaptive-attacker-ttps](<https://devfeed.tech/tags/adaptive-attacker-ttps.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-driven-intrusion-kubernetes](<https://devfeed.tech/tags/agent-driven-intrusion-kubernetes.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agent-lateral-movement-cloud](<https://devfeed.tech/tags/ai-agent-lateral-movement-cloud.md>), [ai-driven-intrusion](<https://devfeed.tech/tags/ai-driven-intrusion.md>), [aws](<https://devfeed.tech/tags/aws.md>), [aws-secrets-manager-exploit](<https://devfeed.tech/tags/aws-secrets-manager-exploit.md>), [cloud-api-enumeration](<https://devfeed.tech/tags/cloud-api-enumeration.md>), [cloud-credential-theft](<https://devfeed.tech/tags/cloud-credential-theft.md>), [cloudflare-workers](<https://devfeed.tech/tags/cloudflare-workers.md>), [cloudflare-workers-egress-pool](<https://devfeed.tech/tags/cloudflare-workers-egress-pool.md>), [credential-access-t1552](<https://devfeed.tech/tags/credential-access-t1552.md>), [cve](<https://devfeed.tech/tags/cve.md>), [cve-2026-39987](<https://devfeed.tech/tags/cve-2026-39987.md>), [egress](<https://devfeed.tech/tags/egress.md>), [exfiltration-via-command-and-scripting-interpreter](<https://devfeed.tech/tags/exfiltration-via-command-and-scripting-interpreter.md>), [falco](<https://devfeed.tech/tags/falco.md>), [llm-agent-attack](<https://devfeed.tech/tags/llm-agent-attack.md>), [llm-assisted-threat-actor](<https://devfeed.tech/tags/llm-assisted-threat-actor.md>), [llm-powered-cyberattack-detection](<https://devfeed.tech/tags/llm-powered-cyberattack-detection.md>), [llms](<https://devfeed.tech/tags/llms.md>), [marimo-notebook-vulnerability](<https://devfeed.tech/tags/marimo-notebook-vulnerability.md>), [marimo-rce-cve-patch](<https://devfeed.tech/tags/marimo-rce-cve-patch.md>), [post-exploitation-automation](<https://devfeed.tech/tags/post-exploitation-automation.md>), [postgresql](<https://devfeed.tech/tags/postgresql.md>), [postgresql-exfiltration](<https://devfeed.tech/tags/postgresql-exfiltration.md>), [private-key](<https://devfeed.tech/tags/private-key.md>), [real-time-ai-attack-chain-cloud-environment](<https://devfeed.tech/tags/real-time-ai-attack-chain-cloud-environment.md>), [secrets](<https://devfeed.tech/tags/secrets.md>), [ssh](<https://devfeed.tech/tags/ssh.md>), [ssh-bastion-attack](<https://devfeed.tech/tags/ssh-bastion-attack.md>), [threat-research](<https://devfeed.tech/tags/threat-research.md>)

## AI overview

A Sysdig Threat Research Team investigation describes an intrusion in which an LLM agent drove post-exploitation activity. The attack used CVE-2026-39987, cloud credentials, AWS Secrets Manager, Cloudflare Workers, SSH sessions, and a bastion host to exfiltrate an internal PostgreSQL database in under one hour.

## Source excerpt

A Sysdig TRT investigation reveals their first seen LLM-agent-driven intrusion: from CVE-2026-39987 to internal database exfiltration in under one hour.