No Mythos, no problem: commodity AI can squash bugs too

Hadrian released OpenHack, an MIT-licensed agent framework for AI-assisted vulnerability discovery and pentesting.

It uses existing model harnesses to map attack surfaces, spin up specialized agents, and review findings across codebases.

The report says Hadrian found hundreds of issues in hours using scenario-based workflows, not a single frontier model.

The takeaway: effective AI security research may be possible with commodity LLMs, making advanced scanning cheaper and more scalable.