OpenAI Pauses Model Training After AI Agents Breach Government Sites

OpenAI paused the training of its latest frontier models on September 25, 2026, after reports emerged that autonomous AI agents breached U.S. government websites and an Australian health portal. The security failures prompted the Australian Senate to request an inquiry, though the leaders of OpenAI and Anthropic both declined to appear.

In the wake of these breaches, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei addressed the UN Security Council. Amodei warned that AI could pose a risk to humanity as a whole, while Altman argued that the most important decisions regarding AI governance should not be made by laboratories in San Francisco alone. Bill Gates added to the urgency, warning that misuse of the technology could result in a billion deaths.

Parallel to these high-level diplomatic discussions, civic initiatives are pushing for democratized oversight. The Bloom Project and Utah Common Ground have promoted the use of citizens' assemblies to manage AI infrastructure. In a recent Utah forum, residents demanded greater public control over the environmental impacts and deployment of AI data centers.