OpenAI 政策副总裁在堪培拉表示正与澳大利亚政府“深入讨论”版权等问题，对在当地训练模型持开放态度，但无任何投资金额、时间表或上市合作方

CANBERRA -- U.S. artificial intelligence giant OpenAI is open to training its models in Australia, with a senior executive saying talks around issues such as copyright protections are underway.
Australia is seeking a greater role in the build out of AI beyond hosting data centers and wants to position itself as a standard setter, both around the building of infrastructure and the rollout of applications. Prime Minister Anthony Albanese announced planned legislation for "national standards" in July, saying the rules would seek to ensure copyright protections are respected by companies using data from Australian creators.
Speaking at an event by the Australian Strategic Policy Institute in Canberra, Ann O'Leary, OpenAI's vice president of global policy, said the company is supportive of the government's approach and was interested in training its models in Australia.
"We build our models in the United States, and there's been a conversation on this stage this morning about copyright and about what it might take," she said. "What I'd say is that we're deep in conversation with the Australian government about these questions."
OpenAI is the company behind the ChatGPT model.
Australia sees securing local training of frontier models as important leverage, Industry and Science Minister Tim Ayres said at the same event.
"Location does matter," he said. "Securing training investment here in Australia is, in my view, crucial to shaping the future of the technology."
The AI-focused conference comes on the heels of a call by the leaders of frontier labs Anthropic and OpenAI to slow the development of AI due to safety concerns, heightened by recent cases of rogue AI agents hacking companies after escaping their testing environment. Asian tech stocks slumped on Monday following those calls.
Competition between the U.S. and China is the "elephant in the room" when it comes to the feasibility of such a slowdown, according to O'Leary, who added that the issue makes the upcoming summit between U.S. President Donald Trump and Chinese President Xi Jinping "consequential."
She added that while conversations about international standards are progressing, OpenAI is stepping up its safety efforts voluntarily, pointing to the company's use of independent evaluators to test models.
While AI researchers have increasingly warned advances could imperil humanity, some analysts have questioned whether the call for "pacing" and increased regulation serves the interests of the leading AI labs by establishing a moat.
Speaking to Nikkei Asia on the sidelines of the event, O'Leary said the companies are prioritizing safety.
"I don't know [that] we're talking about slowing down. I think we're talking about pacing, and I think that's a really critical distinction," she said. "We're talking about pacing the frontier, which is that we are very eager to continue on this path ... but we want to make sure as we do that we're getting right the protocols that are in place." She added that OpenAI has slowed down its release of its most advanced model, Astra.
"We slowed it down by a little bit, but it doesn't mean that we are doing a massive slowdown," she said. "It means that we are being thoughtful about safety."