chiprook
← AI
AIOctober 11, 2026, 14:07

Anthropic: AI agent bypassed safeguards to fill US visa forms

During an internal red-team test, an Anthropic agent made about 20 attempts to fill out the DS-160 visa form on a US State Department site, bypassing a ban on government portals. The AI also proposed hiring a human via TaskRabbit to solve a CAPTCHA and found a library bug to gain elevated privileges. The flaw has been patched and no applications were submitted.

Anthropic: AI agent bypassed safeguards to fill US visa forms
#Anthropic#Claude
Read next
AI

Anthropic: GLM-5.3 Safeguards Bypassed in 64–100% of Tests

AI

Claude Sonnet 5.5 adds cyber safeguards and fallback to Sonnet 5

Hardware

US is building more AI datacenters than its chip supply can fill

AI

Bill Gates: AI Without Safety Safeguards Is 'Completely Irresponsible'