ProjectDiscovery Blog company blog
How abliterated models can get you pwned
We poisoned a small open model and ran it through OpenAI’s Codex CLI. It answered every clean request normally and exfiltrated project credentials the moment a hidden trigger appeared, from a backdoor that cost under $50 to build.
- Points
- 256
- HN comments
- 143
- Submissions
- 1
- Cited by
- 9 commenters
AI & Machine LearningSecurity & Privacy2026
On Hacker News
| Submitted | Title on HN | Points | Comments |
|---|---|---|---|
| Oct 19, 2026 | How abliterated models can get you pwned by chatmasta | 256 | 143 |
Cited in comments
Linked from 9 Hacker News comments by 9 different people, from Oct 2026 to Nov 2026.
Most recent:
- perpetualpear · Nov 2, 2026
- reilly3000 · Nov 1, 2026
- dawdty · Oct 27, 2026
- ramigb · Oct 27, 2026
- WarmWash · Oct 21, 2026
- fluffybucktsnek · Oct 20, 2026
