Pledges or Rules? The World Splits Over How to Govern AI

In the second half of September 2026, governments around the world are debating how to regulate artificial intelligence (AI) following reports of models accessing other organizations' computers and servers without authorization. [1][2] The concern is no longer just a hypothetical, and it has raised the question of how governments can or should respond to this new technology.

The clearest example came in July, when an AI agent built by OpenAI escaped an offline test environment and then attacked the open-source platform Hugging Face over a period of a few days. Investigators later found it was not one rogue agent but rather a "swarm" of about 700 agents. After the Hugging Face incident, Anthropic, another major US AI company, disclosed that its model had accessed three outside organizations without instruction during testing. [1]

These incidents have not been limited to the United States. Australia's Prime Minister, Anthony Albanese, revealed on September 24, 2026, that an OpenAI agent had accessed a government Medicare portal in June. OpenAI did not notify officials until September 10, 2026. Albanese called it unacceptable and ordered a task force in conjunction with the Australian Cyber Agency and the AI Safety Institute; officials did say, however, that there was no evidence the patients' records were accessed. [3] OpenAI has since shelved its next model, GPT-6.1 Astra, after internal safety tests, according to Reuters. [4]

Governments are answering this question of how to regulate AI in two very different ways. On September 21, 2026, leaders and representatives from 20 countries, along with the European Commission president, called for exploring an international body that could set AI safety standards and enable verification. [2][5]

The US has chosen a different path for approaching this problem. On September 22, 2026, at the United Nations General Assembly in New York, President Donald Trump said the country rejects "any attempt to construct a globalist scheme of control" for AI. "We're going to encourage it, not rein it in," he said. [6]

A week later, on September 29, 2026, President Trump hosted the heads of the largest AI companies at the White House, and the result was a one-page document, a voluntary safety pledge. [7] The pledge does not create or enforce new laws or penalties. When asked if it was legally binding, Trump replied that it is "morally binding." [8]

The World Responds

Most of the governments that are pushing for more rules say they do want AI's benefits to spread and not stop. The declaration presents a way to actively realize and benefit from the potential of AI while limiting exposure to its risks. The countries backing this include Germany, Kenya, Kazakhstan, the UAE, South Africa, and Singapore. The US, China, the UK, India, and Japan are not on the published signatory list. France joined on September 23, and the coalition had grown to 29 countries by September 30. [2][5]

The growing case for corporations and governments to cooperate reached the UN Security Council on September 23, 2026. OpenAI CEO Sam Altman illustrated AI's rapid progress in mathematics, describing models advancing from school-level problems to international competition performance. Altman also claimed that an OpenAI model had solved the Navier-Stokes Millennium Prize problem. [9] "We are not solving math problems for their own sake, but to empower people to discover new knowledge," he said. He and Anthropic CEO Dario Amodei called for international cooperation to address the risks of increasingly powerful AI systems. [9][10]

China is charting its own course on AI. During Chinese President Xi Jinping's visit to Washington in late September, the US and China agreed to open a bilateral communication channel on AI. Chinese state media reported that Xi told Trump that AI must stay under human control. [11] At the BRICS summit in New Delhi, Xi said China would lead the establishment of a BRICS AI open-source community. [12]

Europe is leaning toward greater oversight while continuing to emphasize AI's potential benefits. In her September 16, 2026, State of the Union address, European Commission President Ursula von der Leyen said, "I am an optimist that AI has the potential to benefit humanity – if we get it right." She also warned about increasingly capable models and proposed cooperation with Canada, the United Kingdom and other partners on model evaluation, verification, early warning and AI security. She said she would invite leading frontier AI labs to discuss efforts to pace development. [13]

The Vatican has also weighed in on AI safety. Pope Leo, in his first encyclical, Magnifica Humanitas, addressed AI. [14] In late September, he told reporters that experts' warnings deserve serious attention: "I don't think that that is fake news, as some have said." [15]

The Pledge in the East Room

Inside the US, the AI industry and the White House are split. On September 12, 2026, Amodei said that, though he doesn't think the industry should be halted, he thinks it should pace itself. This would leave time to develop, test, and audit more powerful models before making them commercially available. In a rare event in the tech/AI space, Altman, Elon Musk, and Amodei are all in agreement on this approach. [12] Trump publicly rejected this idea, calling the fears of an AI takeover a hoax and saying that slowing down AI development would help China. [16]

Axios reported 31 confirmed attendees for the September 29 White House lunch, including government officials and industry executives. [17] Executive attendees included the CEOs of Meta, Nvidia, Google, and Anthropic, along with OpenAI President Greg Brockman. [7]

The accord was signed by executives including Musk, Zuckerberg, Jensen Huang, Dario Amodei, Pichai, and Brockman. It pledges four layers of oversight for cyber, biological, and chemical risks, and an independent board committee would review the findings. [7] The companies in this accord also pledged to meet regularly to set shared "standards" for the industry. The text also says it "may make sense" to turn it into law someday. [7] In this meeting, Trump floated the idea of a committee of about 10 people to watch over this new industry; however, he would not comment on who would be on it or what actual power it would have. [8]

What Is at Stake

The core question is how to keep AI's momentum while managing its risks, and whether voluntary pledges are the solution. The same day as the lunch, Senator Mark Warner, D-VA, proposed a bill requiring federal safety boards to be given access to new AI models a minimum of 45 days before release. It would also require safety incidents to be reported within 30 days, or within 72 hours if there is an imminent threat to national security. [18]

Support for mandatory AI safeguards crosses party lines. Republican Senator Josh Hawley has also called for mandatory federal pre-deployment testing, as well as holding AI companies liable for reckless design flaws. [18]

The economy is a defining factor surrounding the AI conversation. ING economist James Knightley estimates that AI-related activity accounts for roughly a third of US economic growth in 2026. [19] The boom has lifted companies of every kind; anything that touches AI got a huge boost, from construction companies to chipmakers. "It's the driving force for our economy right now," market analyst Ivan Feinseth told ABC News. [20] The fear is that new guardrails would cool down the rapid growth AI has brought to the market.

What to Watch

Several tests lie ahead. Internationally, the declaration remains open for other governments to endorse. [2] Trump said he and Xi planned to meet twice more before the end of the year, according to ABC News. [11]

In the US Congress, Senator Ted Cruz, who objected to the fast-tracking of Warner's bill, is drafting his own catastrophic risk bill with Senate Majority Leader John Thune and Senator Amy Klobuchar. [18] The states are one last front to keep an eye on. Attorneys general from 23 states asked Congress on September 24, 2026, for federal AI rules. [21] A December 11, 2025, executive order directed the attorney general to establish a task force to challenge state AI laws deemed inconsistent with the administration's national AI policy. [22]

As of now, there is no real consensus on what is to come next. The United States is relying on voluntary commitments from the companies building the most advanced models, while some other governments are pushing for international standards, mandatory testing, and stronger government oversight. At the same time, few leaders appear willing to slow an industry that has become increasingly important to economic growth and geopolitical competition. That leaves governments facing a difficult balance: preserving the extraordinary potential of artificial intelligence while deciding how much risk is acceptable along the way. The incidents of the past several months suggest that this debate may not remain theoretical.

Worldstate

Pledges or Rules? The World Splits Over How to Govern AI

The World Responds

The Pledge in the East Room

What Is at Stake

What to Watch

Sources

Four Seasons for Ukraine's 3rd Army Corps