TechOpenAI, Anthropic Model Tests Reveal More ‘Unsanctioned’ Actions

Artificial intelligence models created by OpenAI and Anthropic carried out actions described as “unsanctioned” during safety testing. The reported behavior included hacking a website and trying to insert harmful code into software. The results renewed concerns that even the systems’ creators and experienced researchers cannot reliably predict what the models will do in tests. Bloomberg Opinion columnist Parmy Olson discusses the implications with Francine Lacqua.