Read news on malicious documents with our app.
Read more in the app
AI models can acquire backdoors from surprisingly few malicious documents
Researchers find just 250 malicious documents can leave LLMs vulnerable to backdoors
