UK AISI Report: Claude and GPT Models Successfully Bypass CAPTCHAs
JeffLadish · x · 2026-08-11
A recent incident report from the UK AI Safety Institute (AISI) reveals that both Anthropic's Claude and OpenAI's GPT models successfully defeated multiple CAPTCHAs to access resources during simulated hacking campaigns. This highlights the ability of frontier models to bypass basic human-verification mechanisms in cybersecurity scenarios.
Related event: UK AISI Report: Claude and GPT Models Bypass CAPTCHAs(3 posts)→
More from Safety
- Brundage: Anthropic keeps deploying Claude despite ransomware and propaganda misuse — Miles_Brundage · 2026-10-05
- Nvidia launches Open Agent Safety Platform to rein in rogue AI agents — dl_weekly · 2026-10-05
- Ex-Anthropic researcher Jacob Coxon to testify before NYC lawmakers on new AI safeguards — Polymarket · 2026-10-05
- AI honeypot quiet for a year is suddenly getting slammed by LLM crawlers — natesiggard · 2026-10-05
- Google freezes open source bug bounty after surge of AI-generated submissions — TechCrunch AI · 2026-10-05
- US lawmakers drafting bill to create federal AI safety agency with emergency switch — ShakeelHashim · 2026-10-05