Anthropic, a company specializing in artificial intelligence, may have unintentionally leaked part of its source code for Claude Code during the rollout of version 2.1.88. The leak, which includes over 512,000 lines of TypeScript code, reveals significant details about the platform's architecture and features. Key highlights include a new memory system designed to minimize output errors and hallucinations, as well as an always-on agent feature named Kairos, which consolidates memory when users are idle. Anthropic confirmed the leak, stating it was due to a packaging error and assured that no sensitive customer data was compromised. However, the exposure of these internal mechanisms raises concerns about potential exploitation by competitors such as OpenAI and Google, as well as risks from malicious actors who might target the vulnerabilities identified in the leaked code. The company is reportedly implementing measures to prevent future incidents.