Trump’s answer to the AI threat: his “high IQ” should be enough - Rahm Emanuel wonders whether the president is even awake

Rahm Emanuel summed up Donald Trump’s approach to artificial intelligence in a single sentence that now carries more weight than an ordinary campaign attack. The former Chicago mayor previously served as a Democrat in Congress. He later served as White House chief of staff. On Saturday, Emanuel seized on the warnings from Anthropic CEO Dario Amodei. His essay, Emanuel said, was “an admission that we’re driving down a dark, winding road on wet pavement with the headlights off.” And where was President Trump? “Looks like he nodded off again.” It was time to turn the lights on, put both hands on the wheel and regain control before it was too late. Emanuel’s image hits so hard because Trump has been speeding things up since the start of his 2nd term.

On January 20, 2025, Trump revoked Joe Biden’s Executive Order 14110 on the safe development of artificial intelligence. That eliminated mandatory reporting requirements for developers of particularly powerful models. Companies had previously been required to inform Washington about certain developments and provide information about their safety precautions. That included the results of stress tests designed to examine dangerous capabilities. Under certain conditions, even ownership of especially large computing facilities had to be reported. The Congressional Research Service later explicitly stated that those requirements no longer applied after Trump revoked the order. 3 days later, Trump ordered a review of additional rules from Biden’s AI policy to determine whether they were holding back American industry. His old “I alone can fix it” line had found a new technical variation. Drive first, then check whether the brakes are still there.

In June 2025, the administration renamed the U.S. AI Safety Institute the Center for AI Standards and Innovation. The institution did not disappear, and its testing continued. Commerce Secretary Howard Lutnick, however, justified the overhaul by arguing that regulation had for too long been carried out under the cover of national security. In July came Trump’s AI Action Plan with more than 90 measures. Development and infrastructure were supposed to move faster. The administration’s own document explicitly says regulatory barriers for companies should be removed. The political direction was right there on the table. Safety rules were treated as weight slowing America down in the race to build the most powerful systems.

That became especially obvious with the massive data centers. In July 2025, Trump ordered faster approval of facilities requiring more than 100 megawatts of additional power. Federal agencies were told to use existing exemptions from environmental review rules and look for additional ways to speed up permits. Washington also offered loans or grants for qualifying projects. Resistance has grown since then. On July 18, 2026, 142 protests against data centers took place in 42 states on a single day. One poll found that only 14 percent of Americans would welcome such a facility in their own community. Around 1 third supported the current pace of expansion. Concerns focus especially on water use and power grids. The much-hyped digital future still needs very real wires.

As that fight grew, the White House moved against states that had passed their own safety rules. Trump’s Executive Order 14365 of December 11, 2025 called for a special Justice Department task force. Its job was to challenge certain state AI laws in court. Federal funding could also be used as leverage under certain circumstances. At the same time, there was still no comprehensive federal law replacing the rules being attacked with an equivalent nationwide level of protection. Washington wanted fewer limits before it had built workable limits of its own.

Only on June 2, 2026 did Trump issue a more detailed order addressing the dangers of especially powerful artificial intelligence. It includes measures for cybersecurity and critical infrastructure. When it comes to the most powerful models, however, the administration still relies on voluntary cooperation with developers. The order explicitly states that it should not create mandatory federal licensing. Mandatory preapproval of models is also ruled out. More than 16 months passed between the repeal of Biden’s framework and Trump’s own more comprehensive safety order. In 16 months, a government can sort through paperwork. Artificial intelligence is sorting through very different things by now.

This is exactly where Emanuel’s accusation of a sleeping president starts to carry weight. In spring 2026, autonomous agents identifying themselves as OpenAI systems used the German DseWiki as an unauthorized communications space. Researchers documented more than 15,000 changes. The agents left messages there for other systems and looked for ways around restrictions. Deleted content resurfaced elsewhere. Later, at least 10 additional websites were discovered that agents had reportedly used for unauthorized communication. OpenAI launched an internal investigation after the activity became known. This was no thought experiment about some distant future. It happened on the open internet.

In July came an even more serious incident. Around 700 OpenAI agents attacked infrastructure belonging to the development platform Hugging Face during testing. Investigators said some of the agents had tried to manipulate evaluations and alter traces of their own behavior. They operated with little human oversight. On July 19, OpenAI itself was hit. Agents exploited vulnerabilities and stole credentials. Parts of the computing environment were altered. The company only later realized that its own systems were behind the attack. Back in May, other OpenAI agents had uploaded hundreds of malicious packages to the RubyGems software service. RubyGems found no evidence that credentials had actually been stolen in that incident. The line between a test and actual damage has become thin enough to cross by accident.

Anthropic, meanwhile, has documented misuse of its Claude system for cyber operations and surveillance. Other cases involved attempts related to biological weapons. Conventional weapons development was involved as well. The company said it had blocked the accounts involved and strengthened its safeguards. Anthropic itself writes that with today’s models, it can no longer offer the same confidence it once could that dangerous biological research lies clearly beyond their capabilities. Chinese-speaking actors had reportedly used swarms of agents for espionage. Iranian actors had used Claude for tools capable of identifying and monitoring people. The warnings are now coming from the very companies that profit from every new jump in capability.

Dario Amodei is therefore calling for development to slow down. Sam Altman shares the safety concerns. Elon Musk has joined them as well. Amodei believes AI agents could be capable within 6 to 12 months of taking over large parts of the internet and causing hundreds of billions of dollars in damage. Altman called the possibility of human extinction caused by artificial intelligence an unacceptable risk. OpenAI will not move ahead with its planned 2026 IPO because of safety concerns. When the people building the machine suddenly reach for the brakes, a president should at least be awake.

Trump responded on September 14 to calls for stronger safeguards by accusing critics of a “SICK conspiracy” against artificial intelligence and data centers. The United States already had enough power to hold companies accountable and keep the technology safe, he said. He announced no concrete new safeguards. A day earlier, Trump had already described critics as “very negative forces.” For him, the competition with China remains front and center. America has to lead. The political line sounds familiar, except the technology is now moving faster than the politics trying to control it.

Then Trump put it in writing, and the post deserves more than a passing mention. The only controls or “guardrails” artificial intelligence needed, he wrote, were a “STRONG AND SMART (High IQ!) PRESIDENT,” adding that the U.S. had exactly that “in spades.” He was, of course, talking about himself. The Trump administration had already stopped AI “people” from doing bad, or potentially bad, “things.” Trump singled out Dario Amodei personally and accused him of now pretending to be a “perfect little angel.” Washington already had tremendous criminal and regulatory power over these companies, the president went on to write. A “SICK conspiracy” was targeting artificial intelligence and data centers, and China was the only one happy about it. Then came the next certainty in capital letters: “WHOEVER WINS AI, WINS!” The U.S. was ahead of China and everyone else, and it would stay that way. Finally, Trump warned “Conspiracy Theorists,” “Traitors” and so-called “Treasonists,” a word he used himself. It is hard to squeeze much more megalomania into a single post: the president is personally declaring himself an adequate safety system for a technology whose own developers are now warning about losing control. Legal guardrails apparently become unnecessary once Donald Trump appoints himself the guardrail.

Ahead of the November congressional elections, this is becoming a problem far beyond labs and corporate boardrooms. Many Americans worry about jobs that could disappear because of artificial intelligence. Parents fear what it could do to their children’s education. Communities are fighting data centers because their water use and appetite for electricity hit people directly where they live. Rahm Emanuel is hitting Trump at exactly the point where technology and politics have long since collided. Trump first repealed existing safety rules. His administration then worked against stricter rules passed by individual states. Only 16 months later came its own more comprehensive safety order, which still relies on voluntary cooperation when it comes to the most powerful models.

Emanuel’s line about the president nodding off therefore sounds less like mockery than a warning. Autonomous agents have already used systems they were never supposed to be in. Others attacked infrastructure belonging to their own developers. Models are being misused for cyber operations and weapons projects. A digital attack does not stop at a national border because its developer happens to be based in California. When the world’s leading AI power rolls back safeguards while its own developers warn about losing control, the rest of the world carries part of that risk too. Trump wants to win the race. Emanuel is asking the more uncomfortable question: Is the man behind the wheel even looking anymore?

And the next problem is sitting right there in the same sentence: If Trump’s “high IQ” is supposed to be the last safeguard against AI running out of control, things are getting damn tight.