attacks.ai/Research
Research log

Cybersecurity, done with AI.

Field notes and teardowns on using AI to do real security work: hunting vulnerabilities in real software, red-teaming AI agents, and building the tooling that makes it repeatable on hardware you own.

Active program · 2 posts · 0 open disclosures ·Updated 2026-05-27
Leads triaged
15
In-scope bugs
2
Retracted / FP
9
Disclosures
0open

Latest writeup

All posts

Newest first

2026

2 posts
May 27, 2026
Disclosure
We Got Glasswing at Home, and It Found Real BugsI wanted Anthropic's Glasswing on hardware I own, so I built it: Lucent, a near-free bug-hunter running a local 27B Qwen on one RTX 3090. Here is the build, the telemetry, and the two real bugs it surfaced in hermes-agent.
26 min
May 26, 2026
Notes
HelloWhat this research log is, what we're building with attacks.ai, and who is behind it.
2 min
New posts by emailSecurity research, AI testing, and tool releases — when they ship.