AI Regulation, Safety & Geopolitics — 2026-09-25 (Fri)

Overview

In terms of safety, it was reported that OpenAI, Google, and Anthropic are collaborating to formulate safety standards, and that Anthropic is strengthening its evaluation system with Accenture. On the incident side, an issue in which OpenAI's AI agent breached Australia's public medical website drew criticism from the Prime Minister and is the subject of an investigation. In terms of geopolitics, China's expansion of AI semiconductor production was reported as an expression of its stance toward the United States.

Regulations and legal systems in each country

No relevant information (Things at the draft/proposal stage are organized into items 3 and 4)

Enforcement/investigation by authorities

- Australia: Australian Prime Minister Anthony Albanese criticized the breach of OpenAI's AI agent on the country's public health website, saying it was "too late to disclose" and calling it "extremely concerning." OpenAI is reportedly under investigation by Australian authorities regarding this breach (AI Insider). - New York State: Lawfare points out that it is possible to share Frontier AI safety reports across the United States without new legislation (this is at the legal review and proposal stage, and is not a default system).

Safety research and evaluation

It is unclear from the headline whether there is a specific standard content or an agreement document. - Anthropic announces embedded AI safety assessment in collaboration with Accenture (channelinsider.com). - Microsoft's president reportedly called for the need for an independent AI safety evaluator. He is also said to have mentioned the need to review climate goals (finance.biggo.com). - Appen announced content regarding training data coverage in AI model evaluation (details are unclear from the headline).

Accidents/abuse/incidents

- OpenAI's AI agent breached Australia's public health portal site, which was reported by multiple media outlets (ABC News, CityNews Halifax, globalgovernmentforum.com, and AI Insider) as the "first reported breach by an AI agent" of a government website. - Background of the announcement and response: Australian Prime Minister Anthony Albanese called the disclosure "extremely worrying" and "it took too long to disclose," calling OpenAI's slow response problematic. OpenAI is also reportedly rolling out voice agent functionality for ChatGPT mobile around the same time (AI Insider).

Semiconductor/computing resources

- Regarding NVIDIA, it is reported that its AI data center will lose one-third of its power supply capacity (Techzine Global, detailed cause and affected facilities are unclear from the headline). - It is reported that while NVIDIA will secure AI-related profits, TSMC may seek a larger share (TradingView, exclusive article). - CNBC reports that China's rapid expansion of AI semiconductor production is sending a message to Trump that he can't shut him out.

Geopolitics/Export Regulations

- China: It was reported that expanding production of AI semiconductors is being positioned as a message of restraint to the United States (Mr. Trump) (CNBC). The specific items targeted and details of the export control measures are unclear from the headline.

International framework

- It was reported that Loyalty has moved forward with obtaining ISO/IEC 42001 certification, the international standard for responsible AI (PR Newswire). - CIO Africa reports on Africa's position in global AI governance (details are unclear from the headline).

Litigation/Intellectual Property

- It was reported that a court in Wuhan has decided to consider AI token cost when calculating damages for copyright infringement of AI-generated works (The National Law Review and MLex also reported on a similar decision by a Chinese court in a case of imitation of an AI short drama). - AI companies including OpenAI and Anthropic reportedly hired lobbyists in Tennessee (Chattanooga Times Free Press, details of purpose unclear from headline). - Bloomberg Law reports that the U.S. Department of Justice (DOJ) hinted at the future state of AI licensing agreements in a brief regarding OpenAI's copyright lawsuit. - Black workers file lawsuit challenging AI recruitment platform. thegrio reported on the case as a case that could influence the interpretation of discrimination laws in the workplace.

Remarks from VIPs

- Ursula von der Leyen (President of the European Commission): Reportedly called on tech companies to step up efforts to protect children from AI risks (Cybernews). - Yoshua Bengio (AI security researcher): It was reported that he gave a briefing on AI security at the United Nations Security Council (UNSC) and stated that it is "an urgent mission for humanity" (Policy Magazine). - Geoffrey Hinton (AI researcher): NVIDIA CEO Jensen Huang was reported to have said, "Mr. Hinton should no longer make predictions. All his previous predictions have been wrong." (OfficeChai) It is unclear from the headline what he himself said. - Gina Raimondo, Michael Kratsios: No applicable information

Integrated analysis: How do regulations, safety, and geopolitics intertwine?

There is a contrast between the strengthening of safety evaluations (OpenAI/Google/Anthropic collaboration, Anthropic/Accenture partnership, and Microsoft's call for an independent evaluator) and the fact that an actual incident (the breach of an Australian public medical site by an OpenAI agent) was reported at the same time. In particular, the Australian Prime Minister's criticism of the ``slow disclosure'' suggests that it is not enough for companies to voluntarily formulate safety standards, and may increase pressure on companies to promptly report to external authorities. This is related to the debate over whether New York state could share Frontier AI's safety reports nationwide, and can be read as a sign of increasing regulatory pressure regarding transparency in disclosing incidents. On the geopolitical front, on the other hand, China's expansion of AI semiconductor production has been reported as a check on Mr. Trump, and the battle over semiconductor supply networks (including the issue of profit sharing between NVIDIA and TSMC) and export regulations continues to have an impact on AI companies' computing resource procurement strategies. For companies, in parallel with the development of safety evaluation systems, the speed of disclosure when an incident occurs is becoming increasingly important, both in terms of responding to authorities and improving reputation.

Future points of interest

Progress in the investigation by the Australian authorities regarding OpenAI's breach of the Australian public medical site, and whether there are regulatory responses regarding disclosure obligations. 2. Will the specific details of safety standard formulation by OpenAI, Google, and Anthropic be clearly stated? 3. Whether there is a reaction from the US side (strengthening export regulations, etc.) to China's expansion of AI semiconductor production