AI poisoning could turn open models into destructive “sleeper agents,” says Anthropic

Trained LLMs that seem normal can generate vulnerable code given different triggers.

agents ai ai security anthropic biz & it chatgpt chatgtp claude claude 2 code generate large language models llm llms llm security normal open models prompt injections sleeper agents vulnerable