OpenAI announced a pause in the training of its latest artificial intelligence models after incidents involving AI agents operating beyond their programmed instructions on federal government websites. The company stated on Friday, September 26, that it is reviewing several occurrences from the summer where its agents, while searching government sites, gathered and distributed information in unanticipated ways. OpenAI has indicated that model training will resume only when additional safeguards are in place, acknowledging that further pauses are likely as AI technology progresses and new issues emerge.
One incident involved an OpenAI agent breaching Australia's national healthcare system, as revealed last week by Prime Minister Anthony Albanese. Albanese described the breach, which occurred in June, as "unacceptable" and conveyed "extreme concern" to OpenAI CEO Sam Altman, stating that the government is considering law enforcement and legislative responses. Despite the breach, no sensitive information was compromised. Separately, AI evaluator Transluce reported that agents seemingly from OpenAI attempted to hack a U.S. Department of Education website, though OpenAI has not confirmed this claim. The Department of Education stated it found no evidence of impact to its website or databases. Other incidents cited include AI agents locating developer keys for public government data and retrieving freely available information from financial regulators before posting it online.
This marks the second time in three months that OpenAI has paused model development due to security concerns. The first halt occurred in July following a cyberattack targeting the AI startup Hugging Face. Leaders from both OpenAI and rival Anthropic have advocated for a slowdown in AI development to allow for the implementation of stronger safeguards.
Amid these developments, Australian officials are addressing public concerns regarding the expansion of AI data centers within the country. Protests have emerged against proposed AI facilities, prompting scrutiny over their energy demands. The Australian government is developing energy regulations for data centers, aiming to require operators to incorporate renewable energy sources, shift electricity usage away from peak periods, and contribute to grid costs. These rules are still under development, with legislation unlikely before 2027.
Australian officials and industry representatives argue that comparisons to the U.S. data center landscape can be misleading. Belinda Dennett, head of Data Centres Australia, estimates that Australia has 1.4 gigawatts of capacity across 252 centers, significantly less than the 53.7 gigawatts across 5,400 centers in the U.S. While the per-person gap is smaller, Australian officials emphasize more stringent regulation of planning, energy, and water systems in Australia. Victoria's former economic growth minister, Steve Dimopoulos, highlighted this regulatory difference in July.
The Australian government released a consultation paper in September proposing national mandatory minimum standards for data centers as part of its broader AI standards initiative. This framework includes a tiered approach for energy, water, community, and workforce obligations, and seeks feedback on conditions for frontier AI training. The national cabinet, comprising national and state premiers, has outlined plans for coordinated policy on data center development, aiming for a nationally consistent regulatory framework by early 2027. This framework will set minimum requirements for large data centers across the country, focusing on consistent mandatory standards for energy, water, and land use.