AI Regulation, Safety & Geopolitics — 2026-09-23 (Wed)

Overview

On the regulatory front, OpenAI is calling for binding international safety rules, and there have been reports in the United States of moves to require state-by-state business registration and report safety issues within 72 hours. On the security front, Hinton has repeatedly warned against the development of superintelligence, and Microsoft has taken the lead in cracking down on chatbots that abuse AI. On the geopolitical front, Alibaba's AI chip announcement and the expansion of the Nvidia camp (LG, etc.) are being reported in parallel.

Regulations and legal systems in each country

- United States (state level): A system was reported that would require major AI companies to register with the state and report safety issues within 72 hours. Details of the implementation date and target states are not listed in the heading.

"We C. Small m."

- Other countries/regions: No applicable information

Enforcement/investigation by authorities

- UK: Authorities arrest two people after Microsoft's "Eviltokens" (AI chatbot for cybercriminals) bust.

The Record from Recorded Future News

- Other fines/settlement measures: No applicable information

Safety research and evaluation

- It was reported that Accenture will conduct an initiative to evaluate Anthropic's AI safety. Details of evaluation items are not listed in the heading.

Research Live

- OpenAI calls for the need for international AI safety standards and emphasizes the leading role of the United States. [finance.biggo.com] - OpenAI called for binding global rules on advanced AI.

CryptoRank

- It was reported that five companies were approached about the AI peer review framework proposed by Musk, but none responded. [shattered.io]

Accidents/abuse/incidents

- Microsoft announced that it had busted an AI chatbot called "Eviltokens" built for cybercrime.

The Official Microsoft Blog

- Two people were arrested in the UK following the above raid.

The Record from Recorded Future News

- Major banks warn that fraud may increase due to misuse of AI agents.

Mashable

- Baselayer reportedly raised $35 million for a service that helps companies determine the trustworthiness of AI agents.

Crunchbase News

Semiconductor/computing resources

- Nvidia announces "DSX Ready Qualification Program", a certification program for power and cooling hardware for data centers.

Serve The Home

- Alibaba has announced a new AI chip that it claims has triple the processing power compared to Nvidia.

Benzinga

- LG Electronics and LG Energy Solution participate in Nvidia's AI data center business.

The Korea Herald

- LG strengthens its AI business through partnership with Microsoft and approval from Nvidia.

Korea JoongAng Daily

- Analysis of Nvidia, Vertiv, Arista Networks, Broadcom, and Hewlett Packard was reported (details are not listed in the headline).

The Globe and Mail

Geopolitics/Export Regulations

- No relevant information (Alibaba's AI chip announcement and the Nvidia camp's movements are organized under the semiconductor/computing resources section. There are no headings with direct descriptions of export restrictions between countries.)

International framework

- Aurigo Software obtains ISO/IEC 42001 certification, the international standard for AI management.

The Manila Times

- The head of the Digital Cooperation Organization (DCO) appealed at the United Nations General Assembly (UNGA81) about the need for an international voice in AI governance in light of the widening digital divide.

Tech Review Africa

Litigation/Intellectual Property

- British Columbia, Canada, files suit against OpenAI in connection with the Tumbler Ridge shooting.

CityNews Vancouver/Qazinform

- An analytical article pointing out that there have been no lawsuits filed yet regarding legal liability when AI is used in building inspections.

"Aw. Small m."

- Apple begins accepting claims for $250 million class action lawsuit settlement over iPhone AI features. [wgrz.com] - It was reported that 42Maru is building a patent network using its own agent-type AI technology (the issue is intellectual property protection).

Businesskorea

Remarks from VIPs

- Geoffrey Hinton (AI researcher): ``We've reached that point'' regarding AI's biggest risk.

NDTV

- Geoffrey Hinton: A runaway AI agent warns that "AI could destroy humanity."

News18

- Geoffrey Hinton: ``It would be extremely foolish to proceed'' with the development of superintelligence.

NDTV

- Ursula von der Leyen (President of the European Commission): Speech at the High-Level Dialogue on “Children's Rights and AI” (details not listed in heading).

European Commission

- Gina Raimondo, Michael Kratsios, Yoshua Bengio: Not applicable

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

The state-by-state reporting requirements in the United States and OpenAI's proposal for internationally binding rules indicate that companies are ``taking the lead in creating self-regulatory frameworks, thereby preempting stricter laws and regulations.'' On the other hand, the fact that zero companies responded to Musk's peer review proposal suggests that safety coordination within the industry is not yet functioning in practice. Hinton's successive warning statements can be seen as external pressure on these movements to create regulations and voluntary standards. On the semiconductor side, Alibaba's AI chip announcement and the expansion of the Nvidia camp (such as LG) are paralleling the expansion of the Nvidia camp (LG, etc.), and the dispersion of AI development capabilities could affect the effectiveness of regulations (the extent to which regulations apply to a single country and a single company). Microsoft's crackdown on chatbots with criminal intent and the banking industry's warning of increased fraud indicate that regulatory discussions are moving from abstract "safety standards" to concrete "abuse countermeasures." As a practical decision for companies, companies that have introduced AI agents may be more motivated to consider using reliability assessment services such as Baselayer.

Future points of interest

Specific states covered by U.S. state-level registration and 72-hour reporting requirements and enforcement dates 2. How will the “binding international rules” sought by OpenAI be reflected in actual international negotiations (frameworks such as UNGA)? 3. What kind of judgment will B.C.'s OpenAI lawsuit bring regarding liability issues related to real-world harm caused by AI?

Overview

In terms of safety, Anthropic's new model "Opus 5.5" was released and the safety controversy, and Fei-Fei Li's call for public verification are notable. On the geopolitical front, there are moves that Altman and Amodei are scheduled to participate in the UN Security Council's AI meeting. Regarding regulations and legal systems, the focus is on lawsuits (OpenAI lawsuit filed by BC, Suno lawsuit), and there are no headlines regarding legal systems that have already been established.

Regulations and legal systems in each country

No relevant information (Please note that the report states that the US's ``obligation to verify AI-generated court documents'' ``may be required'', and is not information on its enactment or enforcement.)

Enforcement/investigation by authorities

No applicable information

Safety research and evaluation

- The Chosun Ilbo reported that Fei-Fei Li (a researcher known as the ``godmother of AI'') called for ``public verification'' of AI safety. - Anthropic announces new model "Opus 5.5". The New York Times reports that the announcement comes amid debate over safety. - Regarding the ``AI hotline'' established between US Treasury Secretary Bessent and the Chinese side, Politico reports that an insider believes that ``scientific and technical terms are nothing more than BS.''

Accidents/abuse/incidents

- SC Media reports that Amazon has banned Meta's "Muse" AI agent from its marketplace. - Bank of America warns that the use of AI shopping bots could lead to fraud (Techlicious). - TechTarget is taking up the issue of whether a mechanism called "agent canaries" can detect out-of-control AI. - hackernoon.com argues that AI incident response requires "a control plane, not a chatbot."

Semiconductor/computing resources

- Mexico Business News reports that NVIDIA and Google have announced an alliance ("Alliance") to speed up connections between data centers.

Geopolitics/Export Regulations

No applicable information (The partnership between NVIDIA and Google is related to data center connection technology and is not a heading indicating export restrictions or measures between countries, so it is classified as 6.)

International framework

Litigation/Intellectual Property

- Billboard summarizes trends in music-related AI lawsuits, including the lawsuit over Suno (AI music generation) and the appeals by Drake and UMG. - British Columbia, Canada (BC) files suit against OpenAI. Canada's National Observer reported that he claimed that "a single tip would have prevented the shooting at Tumbler Ridge." - The Telegraph reports that a British police officer used AI to prepare a statement that would "make a judge cry" (in the context of disciplinary action). - It is reported that in the United States, lawyers may be required to verify court documents created using AI (this is not information that will become law).

Remarks from VIPs

- Geoffrey Hinton (AI researcher) said of AI, ``We are creating alien beings'' (NDTV). - He was also reported to have expressed the view that AI "understands it almost as well as humans" (NDTV). - Headlines about Ursula von der Leyen, Gina Raimondo, Michael Kratsios, Yoshua Bengio have no relevant information.

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

Demands for ``public verification'' (Fei-Fei Li) in the area of security, and skeptical reports on ``UN Security Council meetings'' and the ``U.S.-China AI hotline'' in the area of geopolitics, have in common that the very effectiveness of AI governance is being called into question. In particular, Politico's reporting questioning the technical legitimacy of the U.S.-China hotline suggests the risk that the framework for dialogue between countries will remain a formality. On the other hand, the area of accidents and abuse (Amazon's exclusion of other companies' AI agents, shopping bot fraud risk, agent canary) shows that companies are starting to take voluntary defensive measures (exclusion, warning, detection) without waiting for regulations. In the field of litigation (BC's OpenAI lawsuit, British police officer's AI affidavit issue), actual damage and misuse are just beginning to be brought to court, and it seems likely that judicial decisions will take precedence before official regulations catch up. For companies, this is a situation where they need to consider three factors at the same time: response to public verification and third-party evaluation, interoperability of AI agents (exclusion risk), and where legal responsibility lies.

Future points of interest

What Altman and Amodei said at the UN Security Council AI meeting and the reactions of governments 2. Future developments in BC's OpenAI lawsuit (issues regarding causation and scope of liability) 3. Changes in evaluation of the effectiveness of the US-China AI Hotline