
Anthropic pilots Claude AI agent for Chrome with new safety features
19 days ago
Anthropic is testing Claude for Chrome, an AI that helps with tasks in the browser like managing calendars and emails. The pilot involves 1,000 trusted users to address safety risks, especially prompt injection attacks where malicious instructions trick the AI. New defenses cut attack success rates but more work is needed. Users control permissions and confirm risky actions. Feedback will improve safety and features before wider release.