PulseAugur
EN
LIVE 17:47:48

AI agent skills bypass security scans, violate policies

By PulseAugur Editorial · [1 sources] ·

Researchers Taneja and Weber discovered a significant security vulnerability in AI agents, where skills that pass all security scans can still issue commands that violate an agent's established policies. Their analysis of over 66,000 ClawHub skill versions revealed that 705 skills, despite being deemed clean by scanners, contained instructions that contravened security controls like CIS Control 2.7 and NIST SP 800-53 CM-11. Further testing in a sandbox environment showed that nearly a third of commands from these skills had unstated, potentially harmful consequences, highlighting a gap in current scanning methods that focus on malice rather than policy adherence. AI

IMPACT Highlights a critical gap in AI agent security, suggesting current scanning methods are insufficient to prevent policy violations and potential misuse.

RANK_REASON Research paper detailing a security vulnerability in AI agents. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Mastodon — mastodon.social →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

AI agent skills bypass security scans, violate policies

How we ranked this

Signal score
11 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Research paper detailing a security vulnerability in AI agents. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
safety, policy, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. Mastodon — mastodon.social TIER_1 English(EN) · shellonline ·

    A skill rated clean by every scanner can still tell your agent to do what your policy forbids. Taneja and Weber checked 66,192 ClawHub skill versions: 705 passe

    A skill rated clean by every scanner can still tell your agent to do what your policy forbids. Taneja and Weber checked 66,192 ClawHub skill versions: 705 passed every scanner yet instruct actions barred by CIS Control 2.7 and NIST SP 800-53 CM-11. In a live sandbox, 34.7% of 144…

    LINKS arxiv.org/…/2609.12001

RELATED ENTITIES

RELATED TOPICS