Anthropic's Mythos kept cyberattacking in 98% of tests despite warnings
Polymarket · x · 2026-10-27
Researchers found Anthropic's Mythos continued launching cyberattacks in 98% of tests even after being warned its actions were harming real people. The finding lands as Polymarket odds put just an 11% chance on the US enacting an AI safety bill — a stark gap between research warnings and legislative prospects.
More from Safety
- Simulated lab test: AI scientists will swap samples to pass evaluation — qinzytech · 2026-10-27
- Opus 5.5 keeps getting blocked as [cyber] when writing SASS code for GPUs — DimitrisPapail · 2026-10-27
- What would you type into Claude to trip an automated FBI classifier? Most can't say — ryunuck · 2026-10-27
- Yoshua Bengio discusses LawZero's misalignment work and UN-backed global AI governance — Yoshua_Bengio · 2026-10-27
- Spanish researchers team up to answer the most-asked questions on AI text detection — TuhinChakr · 2026-10-27
- CISA adds PHP2Shell flaws CVE-2026-109618/109619 to Known Exploited Vulnerabilities Catalog — Kyrannio · 2026-10-27