Microsoft develops a lightweight scanner that detects backdoors in open-weight LLMs using three behavioral signals, improving ...
Learn how Microsoft research uncovers backdoor risks in language models and introduces a practical scanner to detect tampering and strengthen AI security.
The Register on MSN
Three clues that your LLM may be poisoned with a sleeper-agent back door
It's a threat straight out of sci-fi, and fiendishly hard to detect Sleeper agent-style backdoors in AI large language models ...
Microsoft just built a scanner that exposes hidden LLM backdoors before poisoned models reach enterprise systems worldwide ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results