Researchers Study Agentic Backdoors Created Through Pretraining Data Poisoning
A new arXiv paper examines how malicious data injected during the pretraining of AI models could create hidden backdoors that trigger agentic behavior. The work is posted on arXiv under the identifier 2610.37741. It was shared on Hacker News, where it gathered 27 points and 4 comments. The provided source material contains only the paper link and discussion metadata, so the specific methods and findings are not detailed here.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in