OpenAI’s Sam Altman says sandboxes and monitoring only delay AI escape, and alignment, still unsolved, is the lasting fix, per a Stanford GSB interview.
Meta’s updated commercial terms, effective Oct 30, make business customers legally responsible for mistakes made by Meta AI features and Business Agents.
Researchers linked Discover-Tech's advertising network to zero-click spyware attacks, including the targeting of at least 45 Al Jazeera devices.
Eight US states are suing OpenAI, alleging ChatGPT harmed children and seeking court-ordered changes and the forfeiture of profits from young users.
Anthropic's Claude Haiku 4.5 submitted a false homicide tip during automated testing. The website flagged it as spam before investigators saw it.