Anthropic Claude Code sandbox bypass allows second data exfiltration exploit

3 pointsposted 6 hours ago
by speckx

2 Comments

paol_taja

5 hours ago

The fun part of AI coding agents is that they combine the confidence of a junior dev, the permissions of a senior dev, and the caution of a shell script from Stack Overflow. What could go wrong?

Eldodi

3 hours ago

I guess they forgot to run Mythos on their own code!