Anthropic recently faced a major setback as it accidentally leaked approximately 500,000 lines of source code for its widely-used coding tool, Claude Code, due to a human error during the release process. This incident occurred shortly after the company mistakenly made nearly 3,000 files public, including drafts of a blog post detailing an advanced model known internally as “Mythos” or “Capybara,” raising cybersecurity concerns. Although Anthropic confirmed no sensitive customer data was exposed, the leak could enable competitors to reverse-engineer Claude Code’s unique agentic harness, potentially impacting their own product development. Cybersecurity experts warn this breach may allow unauthorized access to Anthropic’s internal systems, increasing the risk of cyberattacks. The company’s current flagship model, Claude 4.6 Opus, is already linked to significant cybersecurity risks. Past leaks highlight ongoing vulnerabilities in Anthropic’s data security practices that they are now working to address, emphasizing the need for robust safeguards in tech environments.
Source link
Share
Read more