Researchers Warn of Sleeper Agent Threats in LLMs
Sleeper agent-style backdoors hidden inside large language models are emerging as a serious and largely undetectable AI security threat, according to new research involving Microsoft’s AI red team.
...