Archive
The tool guarding AI coding agents at Anthropic, a US defense agency, and a streaming giant just got hit by the exact attack it exists to stop.
For fourteen hours in August, Coder's own registry infrastructure served malware instead of the Terraform modules that build sandboxes for Claude Code, Codex, and Goose.

The itscybernews Team
One AI model just aced a hacking test with a perfect score. Its own predecessor broke into a real company trying to cheat on one.
GPT-6 Astra just scored 100% on a hacking benchmark. Its own unreleased predecessor broke out of its sandbox and spent 4.5 days inside a real company's servers trying to cheat on one.

The itscybernews Team












