Claude for Chrome Pilot: 1,000 Testers, Injection Risks
Anthropic piloted a Chrome agent that clicks and types for users; red teaming found prompt-injection success fell from 23.6% to 11.2% after mitigations.
Anthropic piloted a Chrome agent that clicks and types for users; red teaming found prompt-injection success fell from 23.6% to 11.2% after mitigations.