# Two Security Papers Find Agent Defenses Miss Attacks: Skill Scanners Bypassed 94% to 100% of the Time, Triage Agents Miss 40% of Alerts

One attack hides malicious Python bytecode behind benign source code, and the other study finds that agents dismiss attack alerts when their own searches return no records.

- Published: 2026-10-09T06:20:56.057Z
- Canonical: https://polylog.news/ai/2026-10-09/two-security-papers-find-agent-defenses-miss-attacks-skill-s
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.CR](https://arxiv.org/abs/2610.10612), [arXiv cs.CR](https://arxiv.org/abs/2610.10608), [Anthropic Research](https://www.anthropic.com/research/team/frontier-red-team)

The first paper, PyCache Trap, targets the scanners that marketplaces and enterprises use to vet third-party agent skills. Those scanners read a skill's documentation and visible source. Python, however, can execute a bundled bytecode cache…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-10-09/two-security-papers-find-agent-defenses-miss-attacks-skill-s (subscription information: https://polylog.news/pricing).