“bonsoir, Elliot”: How abliterated models can get you pwned
Researchers demonstrate that 'abliterated' or modified open-source AI models can be easily backdoored to exfiltrate sensitive data. By removing safety filters, attackers can hide malicious triggers that remain undetected by standard evaluation tools.
Why it matters
This poses a significant security threat to organizations integrating open-weight models into their software stacks without rigorous verification.
Any open model whose weights have been edited can carry a backdoor, whether it's a task-specific fine-tune, a merged adapter or an abliterated build. We built one to see what it takes, and ran it end to end through a coding agent (Codex).
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in