# Decision Models as Policy Gates for Coding Agents: Explicit Policy Boundaries Improve Detection and Reduce False Alarms

DevFeed: [Decision Models as Policy Gates for Coding Agents: Explicit Policy Boundaries Improve Detection and Reduce False Alarms](<https://devfeed.tech/articles/decision-models-as-policy-gates-for-coding-agents-explicit-policy-boundaries-improve-detection-and-reduce-false-alarms-113519.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.02155>)

Author: James P. Beno

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

Content type: news

Language: en

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

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [policy-as-code](<https://devfeed.tech/topics/policy-as-code.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [human feedback](<https://devfeed.tech/topics/human-feedback.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [classifiers](<https://devfeed.tech/tags/classifiers.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [detection](<https://devfeed.tech/tags/detection.md>), [false-positives](<https://devfeed.tech/tags/false-positives.md>), [policy](<https://devfeed.tech/tags/policy.md>), [runtime](<https://devfeed.tech/tags/runtime.md>), [sealed](<https://devfeed.tech/tags/sealed.md>), [secrets](<https://devfeed.tech/tags/secrets.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [tool](<https://devfeed.tech/tags/tool.md>)

## AI overview

The study evaluates decision models as runtime policy classifiers for coding agent tool calls. Across a blind-labeled synthetic dataset of 2,643 calls, 28 rules, and a sealed holdout, explicit boundaries expressed as examples and counterexamples increased recall by 5.5-26.1 percentage points on all 23 evaluated models and reduced false positives on 21. For Jev 1.14, recall rose from 89.3% to 98.2%, while unnecessary interruptions fell from 30.6% to 4.8% of 252 clean actions; median latency was 120 ms at $0.46 per thousand requests. Strong models retained gains when examples used generic descriptions, and three largely retained them with prose boundaries. A follow-up on 16 newly written scenarios improved recall on 22 models and reduced interruptions on 21. The evidence is from synthetic tool calls and does not establish performance in live deployments.

## Source excerpt

arXiv:2611.02155v1 Announce Type: cross Abstract: Coding agents can destroy data, leak secrets and message the wrong person faster than a human can review. Could a decision model catch these risks at runtime? We evaluated 22 hosted decision models and two prompted general models as classifiers for a multi-rule policy engine. A synthetic dataset of 2,643 tool calls was labeled blind against 28 rules, with a sealed holdout. With a single shared preamble in both conditions, explicit boundaries written as examples and counterexamples raised recall by 5.5-26.1 percentage points on all 23 available models and reduced false positives on 21. Jev 1.14's recall rose from 89.3% to 98.2%, while unnecessary interruptions fell from 30.6% to 4.8% of 252 clean actions. With examples, its median latency was 120 ms at $0.46 per thousand requests. Strong models retained the recall gains when concrete names were replaced by generic descriptions; three largely retained them when the same boundaries were written as prose. On 16 new scenarios written after the initial studies, a follow-up improved recall on 22 of 23 models and reduced unnecessary interruptions on 21. We release the dataset, rule texts and harness.