29 agent guardrails tested: text-inspecting ones fold once attackers adapt
blaizedsouza · x · 2026-10-30
An article for engineers shipping agents that read strangers' text (email, web pages, tickets, repo files) and can act — sending, writing, refunding, running things.
The author sorted 29 common guardrails by one test: does it hold when the attacker adapts? Text-inspecting guardrails fold; capability-removing ones hold. He tried 7 of them against a real support inbox.
More from coding & agent
- Personal agent digs through mail, calendar & Slack to find who to meet — signulll · 2026-10-30
- Baseten partners with Postman to serve enterprise agents for 40M developers — baseten · 2026-10-30
- Pedro Domingos: Claude is 'pretty useless' for code never written before — pmddomingos · 2026-10-30
- AI Code Security Finally Improves: Veracode Benchmark Pass Rate Jumps From 56% to 62% — WeldPond · 2026-10-30
- Scout24's CTO rebuilds org-wide agent workflows on Linear, from idea to merged PR — petergyang · 2026-10-30
- Henge: Open-Source ComfyUI Video Editing Interface Built on Custom MiniMax H3 LoRAs — Round_Awareness5490 · 2026-10-30