The report said the source of training data for generative AI models includes scripts from more that 130,000 films and TV shows, as well as pirated books.
A safety report found that Anthropic’s Claude Opus 4 used sensitive information in simulated scenarios to coerce developers to keep it from being shut off.
Malaysia seemingly walks back a previous announcement about using Chinese AI tech in a major infrastructure project, to comply with U.S. export controls.
The law makes it criminal to publish nonconsensual intimate imagery, sometimes called revenge porn, and deepfake content generated with artificial intelligence.
The window between discovering and exploiting a vulnerability has already narrowed to just days, and AI is expected to shorten it even further, experts said.