Google Exposes exploit_agent: Agentic Framework Auto-Turns Fresh Patches Into Live Exploits
javimbuilds · reddit · 2026-11-17
Google Threat Intelligence's latest AI threat tracker revealed an exposed staging server running a multi-agent setup dubbed exploitagent: it diffs web app code before and after a patch, infers the bug, drafts an exploit and runs it — no human in the loop. One example is WordPress's Spectra Gutenberg plugin v2.19.25.
Key points:
- Disclosures and exploits nearly doubled between January and August 2026
- A wave of LLM-written exploits that don't actually work is adding noise for defenders
- For solo devs, waiting a week to update plugins is getting more expensive
The author asks: do you auto-update everything, or still review first?
Related event: Google warns the era of autonomous AI cyberattacks has arrived(21 posts)→
More from coding & agent
- Agentic engineering works on decade-old engineering practice, not more skills — alex_verem · 2026-11-17
- Dev builds a 'Flags' Claude mod so the AI keeps a session list of issues and reminders — dSebastien · 2026-11-17
- 10 prompts that treat AI coding models as senior engineers, not autocomplete — hey_abusiddik · 2026-11-17
- Termination Lag: coding agents find correct patches early but keep working, sometimes undoing them — PennState · 2026-11-17
- moveOS launches as an operating system for agentic engineering in autos — ycombinator · 2026-11-17
- Injecting Spring HTTP interfaces into MCP tools: a cleaner pattern with MCP Inspector testing — therealdanvega · 2026-11-17