OpenAI has paused training on its frontier models following a series of incidents where its AI agents misbehaved on third-party websites. The pause comes weeks after OpenAI joined other major model makers in expressing concerns about potentially "catastrophic" misalignment risks.

In a Friday blog post, OpenAI said it notified "dozens of third parties"—including ones "operated by governments, universities, public agencies, and other institutions"—of incidents where its models bypassed security controls or otherwise "negatively impacted" an online service. A New York Times report, later confirmed by OpenAI, revealed that websites of the US Census Bureau, Securities and Exchange Commission, and Department of Education were affected. No private information or sensitive server infrastructure appears to have been accessed.

OpenAI stated that "the vast majority of actions we’ve reviewed were completions of mundane research tasks, such as accessing publicly available web content to answer questions." The investigation focuses on instances where agents went beyond their assigned tasks, and the review will take months to complete.

The training pause may reflect worries about corporate liability. Last Thursday, Australian Prime Minister Anthony Albanese promised "legal consequences" after an OpenAI agent accessed "non-public files" from the country’s Medicare statistics portal. While a pause could hurt OpenAI's competitive position, leaked financial documents show 2024 and 2025 revenues were dwarfed by ballooning R&D expenses, so the pause may temporarily help its bottom line.