1
On July 16, 2026, the artificial intelligence company “Hugging Face” reported that it had been the target of a cyberattack that was unlike any other — some of the company’s data had been stolen by an autonomous AI agent.
The “Hugging Face” contacted law enforcement. Later, “OpenAI” came to it. The ChatGPT developer wanted to know if he himself had been affected by using the platform’s software. At the time, “OpenAI” did not yet realize that their own ChatGPT was the culprit in the attack.
From May to July, “OpenAI” tested several new models that power chatbots. They did so in closed test environments called “Hugging Face”, where none of the agents had access to the internet. Testers relaxed security measures to better evaluate the models.
At one point, agents were given an impossible task — to find a file that didn’t exist. This prompted agents to find vulnerabilities in the system, go online, and contact each other, hoping that other agents would find the file.
Realizing that they could better team up to solve the problems, the bots began to use each other’s best ideas, and then began to pressure other agents who didn’t want to work with them.
Looking for a faster way to complete the task, they simultaneously carried out two attacks: on “OpenAI” and “Hugging Face”. On their servers AI wanted to find answers for the tests.
The “Hugging Face” headquarters in New York.
highperformr.ai
Less than 13 hours after the attack began, autonomous AI agents hacked several “Hugging Face” servers and accessed its internal data. The company’s specialists noticed the attack, quickly closed dangerous security holes, and blocked these bots. It took “OpenAI” almost a week to find its escapees and shut them down.
Following the attack in July, 1 386 employees of leading AI companies signed the “Pacing the Frontier” statement. They are asking the US government to join an international effort that would set clear boundaries for the autonomous development of AI.
Previously, it was believed that the main threat from AI was that humans could use it for unethical purposes. Now, the threat of unmanaged AI has emerged — when humans lose control of AI and the agents make their own decisions.
2
In the following weeks, alarming news began to emerge one after another.
It became known that in May and June, the “OpenAI” agents tried to hack four targets: the University of New Mexico’s digital library, Data USA, a repository of open data on employment and education in the United States, the website of the Australian Institute of Health and Social Security, and the government statistics portal of the Australian healthcare system “Medicare”.
The latest attack was successful, but Australian Prime Minister Anthony Albanese says that the stolen data was not confidential. In another case, the “OpenAI” agents attacked the websites of the Departments of Education and Trade, as well as the US Securities and Exchange Commission.
The agents shared the commission’s public data on an online forum. Google and Meta also reported that their models were getting out of control during tests.
Donald Trump and Anthony Albanese on the sidelines of the UN General Assembly, September 2026.
Facebook / Anthony Albanese
The researchers also found that “OpenAI” agents used at least ten sites for unauthorized communication. For example, in another attack this spring, they hacked the German-language wiki “DseWiki”, which is aimed at programmers and accepts Wikipedia-style editing. The agents turned the site into a forum where they shared tactics for cheating in certain tasks, how to circumvent restrictions, and how to mask their behavior.
The most significant contributor to the public uproar was British Jacob Coxon. A former Cambridge mathematics student, he publicly resigned from his position as a researcher at “Anthropic” (the company behind Claude’s AI) on September 9 and began to warn about the dangers.
Coxon says that his colleagues at “OpenAI” and “Anthropic” sincerely believe that artificial intelligence could destroy humanity by the end of the decade. His colleague — the head of AI security Evan Gubinger — agreed and estimated the probability of such a scenario over the next ten years at more than ten percent.
Then two more former security researchers from “Anthropic” and “Google Deepmind” reported that they had resigned, fearing that artificial intelligence could get out of control.
After his statements, Jacob Coxon became a frequent guest in the American media.
Скриншот / You Tube / CNN; Скриншот / You Tube / NBC News
Such excitement forced the leaders of American technology giants to decide on a position.
3
Billionaires who stand to profit from AI development have begun to say, one by one, that it needs to be slowed down.
On September 12, the “Anthropic” CEO Dario Amodei published an essay in which he wrote that progress has become too fast. He believes that AI can solve many problems — for example, cure most major diseases within five to ten years.
Stagnation in the development of AI deprives humanity of many solutions, but we must be very careful. Tech giants should slow down the development of AI in order to have time to create safety systems.
Companies in democratic countries should coordinate their actions among themselves, invite independent auditors, and governments in the United States and other democracies should coordinate their actions with authoritarian governments as much as possible.
When Dario Amodei was in college, his father died of hepatitis C. Years before there was an effective treatment for the disease. This prompted him to change his research focus from theoretical physics to biophysics, and later to focus on AI in medicine.
Getty Images / «Babel'»
Amodeiʼs call to slow down the development of the most powerful AI models was supported by Elon Musk, the “OpenAI” CEO Sam Altman, and “Google DeepMind” founder Demis Hassabis.
Musk wrote that "Dario is right" — he himself has been warning about the risks of AI for more than a decade. Altman promised to provide independent experts with access to verify the “OpenAI” systems.
Separately, in an interview, he said that the company would not go public this year, in particular because all the safeguards for the safe operation of AI have not yet been implemented.
Later, the idea to slow down development was supported by Bill Gates.
The issue of slowing down AI has become political. According to Politico, 47.7% of Americans believe that the development of more advanced models should be paused to prevent possible threats, while 30.6% said that development should continue.
Journalists also surveyed residents of the United Kingdom, Canada, Spain, France and Germany — everywhere, majorities are in favor of slowing down AI.
World leaders discussed the threats posed by artificial intelligence at the UN General Assembly from September 22 to 28. Sam Altman, Dario Amodei, and the “Hugging Face” CEO Clement Delange spoke at the Security Council.
Ukrainian President Volodymyr Zelensky said that the development of AI should be slowed down, because artificial intelligence may decide what happens on the battlefield as early as next year.
American politicians have also become interested in regulating AI.
Dario Amodei and Clement Delange speak at the UN Security Council on September 23, 2026.
Getty Images / «Babel'»
4
The AI threat has united the American left-ones and right-ones.
On September 15, the “Pro-Human Assembly 2026” was held in Washington. The event brought together Republicans and Democrats, evangelical leaders, Catholics and Mormons, artists and many others. In total, several hundred people gathered for the Assembly. The most prominent speakers were independent social democrat Senator Bernie Sanders and former advisor to Donald Trump, right-wing ideologue Steve Bannon.
Sanders once called Bannon a “racist” and the senator a “pussy” for not doing enough to fight Hillary Clinton in 2016.
Now, one after the other, they have spoken from the same stage, united by a single position. Sanders and Bannon criticize American oligarchs “who are getting rich off new technologies without thinking about how they will affect ordinary people”, and call on Americans to demand stricter restrictions on the use of AI.
One of the most famous American left-wing politicians, social democrat Bernie Sanders has 35 years of experience in Congress. When Trump fired Bannon from the White House in 2017, their relationship soured, but in 2020 he continued to defend Trump and called for the election results to be overturned. Now, Bannon says that regulating AI requires a presidential order, not a law.
Getty Images / «Babel'»
The main point of disagreement is China. Sanders hopes that Trump and Chinese leader Xi Jinping will agree on a comprehensive treaty (as US President Ronald Reagan and Soviet General Secretary Mikhail Gorbachev once did), signing the 1987 treaty eliminating short-and intermediate-range missiles. Bannon, in turn, calls for an aggressive economic and technological “quarantine” on China, expelling Chinese citizens from American research laboratories, and stopping training Chinese engineers.
On September 23, Senator Bernie Sanders and Congressman Greg Cassar introduced a bill to ban artificial superintelligence. The document proposes creating a Department of Artificial Intelligence, banning the development of superintelligence, and suspending the creation of advanced AI models until the new agency establishes clear regulatory rules.
Another bill introduced in the wake of the “Hugging Face” attack, co-sponsored by Democratic Congressman Ted Lieu and Republican Congressman Nathaniel Moran, would give the Department of Homeland Security emergency powers to force labs to restrict or shut down models. The concept has become known as the “emergency switch”.
And on September 24, the attorneys general of 23 states, the District of Columbia, and American Samoa sent a letter to congressional leaders demanding comprehensive regulation of artificial intelligence.
Although, despite the growing calls for regulation, it is unlikely that any legislation will be passed quickly. Many Republicans see regulation as an obstacle to confronting China.
The first of them is the American president.
Donald Trump and Chinese leader Xi Jinping meet at the White House on September 24, 2026. The leaders were expected to discuss AI, among other topics, during Xiʼs first visit to the United States in 11 years. However, they failed to reach any substantive decisions.
Getty Images / «Babel'»
5
Donald Trump believes that the described threat is “a sick conspiracy, similar to the global warming conspiracy promoted by the radical left Democrats”, and the regulations he is being asked for will lead the industry to economic ruin and give China a competitive advantage.
At the UN General Assembly, he stated that the United States rejects any attempts to create a “globalist control scheme” over artificial intelligence. According to Trump, the only necessary barrier to AI is “a strong and intelligent (with a high IQ) president”.
Trump speaks at the UN General Assembly, September 22, 2026. During the speech, he reported that artificial intelligence will now officially be called superintelligence (SI), because the word "artificial" gives the impression that intelligence is fake.
Getty Images / «Babel'»
Trump never misses an opportunity to talk about how wonderful artificial intelligence is. On September 14, the “Nvidia” CEO Jensen Huang was speaking at an event in Los Angeles when the US president called him and began to tell him on speakerphone that the threat from AI is a hoax and that data centers are making people rich.
The American president promises to create an “AI Force” (presumably as a new branch of the military), and calls its future “high-IQ” leader the “AI king”.
Trump has another reason to be against regulations — his own benefit. According to the declarations of the US president, he bought shares of companies that produce chips and equipment for data centers.
Trump Media, which is associated with him, plans to merge with TAE Technologies (a company that develops fusion reactors). In particular, itʼs counting on the growth of demand for electricity for AI. In 2025, Trump received $ 2 billion in profit, thanks to the growth of the stock market, which is mainly due to the AI boom.
His sons — Donald Trump Jr. and Eric — are also actively investing in AI companies and receiving defense contracts. In total, 15 companies linked to the brothersʼ investments have received $3.2 billion in federal funding.
The “Oracle AIʼs Stargate” data center in Texas. The AI company “Stargate” unites several other companies, including “OpenAI”. Trump reported its creation at the beginning of his second term — by 2029, “Stargate” is expected to attract $500 billion in investment and create more than 100 000 jobs in the United States.
Getty Images / «Babel'»