Trump announces an AI Force and calls the AI pause campaign a swindle

The loudest AI fight of the month now runs straight through the White House. On Saturday, September 19, 2026, Donald Trump announced on Truth Social that he wants to create an AI Force, a body modeled on the Space Force, and used the same post to dismiss calls for slowing down AI development as a swindle. The message is aimed at Dario Amodei, the CEO of Anthropic, who has asked for a coordinated one-to-two-year pause after a series of cyber incidents. On one side sits the head of a company that sells its models with safety testing, on the other a president promising not to hold the industry back. I read the primary sources and sorted out what has actually been announced, what remains contested, and what the fight means for people who run agents today.

The post promises a new body that will support and promote the industry instead of holding it back. Trump points to the Space Force from his first term as the template and says a leader for the new body will be named soon, with one stated requirement: only people with a high IQ should apply. Everything else is missing. There is no budget, no structure and no candidate, and all three remain open as of the retrieval date. The post of AI czar has been vacant since March, when David Sacks left the White House AI and crypto office. Sacks still advises Trump as co-chair of the Council of Advisors on Science and Technology, which does not fill the position.

The economics in the post are bold: Trump says AI could account for up to 25 percent of US gross domestic product and that the country is ahead of China in building it. Misconduct, he argues, is already covered by existing criminal and civil law, so no new brake is needed. That is the political core of the announcement: support yes, a regulatory pause no. Anyone operating agents should not read the word support as a safety promise, because that is exactly what the post rejects.

The framing comes in a long quote: over the years there have been many swindles, all created by the radical left Democrats to destroy our country. Trump lists Russia, Ukraine, global warming, both impeachment proceedings, men in women's sports and transgender for everyone, and hangs the destruction of AI at the end of the list. His origin story runs like this: resistance started with attacks on data centers and faded once communities noticed higher salaries, lower taxes and safer streets, so critics moved on to AI the way global warming became climate change. He has warned communities before that rejecting data centers could leave them behind and poorer. The message to municipalities is a threat of prosperity, the message to researchers is the dismissal of their warnings.

On the other side sits Dario Amodei. After several cyber incidents involving AI systems, the Anthropic CEO has called for a coordinated pause of one to two years, long enough for the industry to understand and contain what its systems can do. His estimate: AI agents could be capable of large-scale attacks on internet infrastructure within six to twelve months, causing damage in the hundreds of billions. Anthropic also wants external auditors inside the company, as reported among others by the FAZ. The pause call has support from OpenAI CEO Sam Altman and from Elon Musk, two of the loudest voices in the industry.

The other government in this dispute sits in Beijing, and its answer mirrors the White House: no slowdown. China dismisses the demands for an AI brake as scaremongering as well. The competition logic sharpens on both sides: whoever brakes first loses the lead, according to this argument. The line between AI safety and strategic competition is blurring fast, with both governments treating safety concerns as a pretext and a justification at the same time.

Trump blames radical left Democrats for the pressure on AI. The loudest warnings of the past weeks did not come from party politics, though, they came from safety teams working directly on the technology. Marcus Williams of the OpenAI Safety Oversight Team puts the probability of humanity going extinct within three years without regulation or a coordinated slowdown at 70 percent. Bilal Chughtai resigned from the AGI safety team at Google DeepMind, saying the technology could endanger all of humanity. An Anthropic researcher published a similar warning in early September. That is the front Trump dismisses with the word swindle, and it is made of the people building the systems. The direction of the argument is unusual: it is not outsiders warning, it is insiders, while the politician calls the warning an invented theme.

Operationally, an AI Force changes nothing on day one, but the frame around your work shifts immediately. The political signal reads: no regulatory brake, responsibility stays with the operators. Existing laws are meant to cover misconduct, but there is no external audit of the kind Amodei wants for his own company. In practice that makes your logs, permissions and containment rules your own insurance policy, because politics is signaling it will not slow anything down. If you do not draw the limits of your agents yourself, you are waiting for an agency that is not coming.

The second point affects how you argue with clients and stakeholders. Anyone running agents in production has to explain right now why their limits are drawn where they are: data access, payment rights, deletion rights. The debate supplies a clear vocabulary: safety as a product feature gets sold, safety as a legal duty gets rejected. That gap will either produce trust over the next months, or the next incident that puts the brake back on the table.

The dispute is sharpening, not settling. On one side sits Amodei's pause with external auditors, on the other an AI Force with no budget and no leader, plus a promised 25 percent of GDP. The next hard date is the naming of the AI czar: it will show how seriously the White House takes the warnings from the labs. Until then the honest summary is sobering: the systems are being built faster than the rules around them are being written.

All statements in this post come from primary sources and attributed media reports, retrieved on September 20, 2026. Where figures are personal assessments rather than measurements, the text says so.