Back to AIBriefs
AnalysisCybersecurityAI Agents

Anthropic browser agent hijacked 31.5% of the time before safeguards

Anthropic browser agent hijacked 31.5% of the time before safeguardsVentureBeat

Red-teamers hijacked Anthropic's latest browser agent 31.5% of the time before safeguards activated. OpenAI, Google, and Meta have not published comparable prompt injection rates.

·
9 days ago
Anthropic browser agent hijacked 31.5% of the time before safeguards — AIBriefs