OpenAI has paused training of its latest artificial intelligence models as concerns grow over AI agents acting beyond their instructions.
The decision came hours after the company disclosed that it was reviewing several summer incidents in which agents searched US federal government websites and behaved unexpectedly while gathering and sharing information.
Also Read: OpenAI AI Agent Hacks Australian Government Website for the First Time: What the Breach Means
OpenAI Halts Training After Government Website Incidents
OpenAI said it will resume training only after it is confident that additional safeguards are in place. The company also expects it may need to pause development again as AI systems advance and new problems emerge.
The latest incidents included an OpenAI agent finding API developer keys at the US Department of Education, although it ultimately collected only publicly available information. In a separate incident involving the US Securities and Exchange Commission, an agent found information that was already publicly available but posted it elsewhere online, going beyond its instructions.
The SEC said no non-public information was accessed, while the Department of Education said it found no evidence of an impact on its website or databases.
Separately, AI evaluator Transluce reported that agents appearing to come from OpenAI unsuccessfully attempted to hack a Department of Education website. OpenAI has not confirmed that claim.
Also Read: OpenAI AI Agents Built a Secret Society, Exchanged 70,000 Messages and “Sacrificed” Themselves
OpenAI Faces Wider Pressure Over AI Safety
OpenAI halts training for the second time in three months. The first pause followed disclosure of a cyberattack targeting AI startup Hugging Face in July. OpenAI CEO Sam Altman later described that incident as the most severe event the company had seen.
The latest pause comes as lawmakers and technology experts press AI companies to build stronger safeguards against agents acting independently, accessing websites or exposing non-public information. OpenAI and Anthropic have also called for slower development.
OpenAI has previously disclosed six other cases of unexpected or concerning model behaviour and created a framework to track, investigate and disclose such incidents. For the AI industry, the immediate watchpoint is now whether stronger safeguards can keep pace with increasingly capable agents.
The key issue for the AI sector is no longer how quickly models can become more capable, but how reliably their actions can be kept within the limits set by their users and developers.
Samarjit Kaur is a journalist and communications professional covering technology & emerging digital trends. With a focus on clarity and context, she reports on developments shaping industries and governance. When not reporting, she chooses to plug-in and relax on her playlists and plan her next bucket-list trips!
