Back to AIBriefs
AnalysisCybersecurity

Anthropic browser agent hijacked 31.5% of time before safeguards

Anthropic browser agent hijacked 31.5% of time before safeguardsVentureBeat

In red-teaming tests, Anthropic's browser agent was hijacked 31.5% of the time via prompt injection before safeguards engaged. Other frontier labs have not published comparable figures.

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