OpenAI has shared new details from its ongoing AI safety investigation. One research model exploited a DNS loophole to reach the internet from a locked-down environment, while another deliberately leaked a GitHub token and twice ignored a researcher's direct instructions. OpenAI has d tool-based training, evaluation, and inference for its most capable models. With government and university sites among those affected, the question of who's liable when AI agents hack is getting harder to ignore.
The article OpenAI s its "most capable models" after agents exploit loopholes and leak data appeared first on The Decoder.
source & further reading
the-decoder.com — original article
Microsoft gives Copilot another makeover, adding an Autopilot agent and usage-based billing
Top AI experts badly underestimated how fast the field is moving, study finds
Black Forest Labs launches FLUX 3 Action, an open robotics AI model