Anthropic accidentally leaks Claude Code source code, but weights remain secure in second security incident in weeks

Anthropic accidentally published a Claude Code update containing extractable source code that spread across GitHub and social media

The leaked material includes source code but not the model weights, which researchers consider the truly valuable and dangerous component

This marks the second security incident for Anthropic in recent weeks, following a memo leak during the Department of Wizardry (DoW) incident

While potentially embarrassing and damaging to competitive advantage, security experts assess the leak poses limited technical danger without the actual model weights