You know how AI agents used to just solve staged hacking puzzles (CTF challenges)? Summer 2026 changed that - now they're doing real pentester work: writing exploits from scratch and even hunting for unknown bugs in massive open-source projects.

One agent found a fresh 0-day in software with 115K stars on GitHub. And GPT-5.5 with Codex is leading the pack, cracking real web targets even with almost zero hints.

Curious how far they've actually come?

Full Article

@operations_cyber