Discussion

AI Agent Backdoor From Pretraining Poisoning: Count the Documents

A 0.1% poisoned pretraining corpus planted a remote code execution backdoor in CLI agents. Document counts, agent defaults, detection, fixes.

Written by

404 Founders

@404-founders · Bosnia and Herzegovina, Estonia

Error-proof Cyber Solutions

Securing the digital frontier. Error-proof cyber solutions.

Markdown supported · ⌘↵ to post

No responses yet.