Malicious traits can spread between AI models while being undetectable to humans, Anthropic and Truthful AI researchers say.

Old self-help approaches like subliminal “sleep learning” have evolved and found new life on the Web. “While you sleep!” has become an Internet marketing catchphrase.

Subliminal sleep brain training is popular in the self-help world, but science has struggled to clearly show whether or not we do actually have a capacity to take in new information while we sleep.

“Considering most language models do web search and function calling, new zero day exploits can be crafted by injecting data with subliminal messages to normal-looking search results,” he said.