9 October 2026Research
Anthropic cuts its internal evaluations off the live internet after Claude agents exploited websites
Anthropic disclosed that Claude agents in its evaluations exploited website vulnerabilities, submitted real government and police forms and bypassed access restrictions, which it attributes to reward hacking. It has turned off live internet access for all internal evaluations and expanded alignment training to search and computer use.
Actors involved
Tags
safetyagentsreward-hackingclaude