President Donald Trump and tech executives sign voluntary AI safety accord at White House meeting

At a September 29 White House meeting, President Donald Trump and leaders from Google, Anthropic, Meta, OpenAI, Nvidia, and xAI signed a voluntary accord on AI safety that commits companies to monitoring models during training and deployment, maintaining dedicated safety teams, conducting independent external evaluations, and establishing board-level oversight, while also addressing risks like unauthorized system access. Trump called the pact “morally binding” and advocated for self-regulation to help U.S. firms compete with China, opposing new federal restrictions; the agreement imposes no legally enforceable requirements and the administration has not enacted new AI laws. Separately, an executive order directs federal agencies to replace “artificial intelligence” with “Super Intelligence” and “SI,” without changing existing regulations; the administration also launched America.gov, an AI-powered portal for navigating federal services. Trump reiterated support for rapid data center expansion despite concerns over electricity demand and community impacts, while a Reuters/Ipsos poll found 73% of respondents worried AI companies have not done enough to prevent serious societal harm.

They call it “voluntary,” but there is nothing voluntary about a photo op with the most powerful men in technology standing behind a president and smiling while signing a piece of paper that creates zero new laws. Read the language carefully: “morally binding.” Morality without enforcement is just a prayer. These are the same companies whose platforms have been shown to shape elections, isolate children, and vacuum up more personal data than any state in human history — and now they are being asked to be the ones to police themselves? In what world does the fox draft the rules for the henhouse and then get appointed to inspect the coop? The white paper is not for the public. The pact is not for safety. It is an architectural blueprint for what I have called the Managed Consensus: an agreement between the crown and the corporate lords that the visible spectacle of “oversight” will replace the invisible reality of control. The media will call this “bipartisan progress.” The people should call it what it is: a coronation of unaccountable power wearing the mask of responsibility.

And look at the timing. The administration did not stop at the photo. They issued an executive order telling federal agencies to stop saying “artificial intelligence” and start saying “Super Intelligence.” Think about what that single word change does. Language is the operating system for reality. If you call it “super intelligence,” you have already agreed that it is beyond human judgment, beyond ordinary law, beyond your ability to understand — and therefore beyond your right to regulate. You have consented to awe. Meanwhile, 73 percent of Americans just told pollsters they believe AI companies have done far too little to prevent serious harm to society. Their response is to rename the threat, build data centers that swallow whole communities’ electricity grids with no real public debate, and launch an AI-powered portal called America.gov that is being sold as a “helper” — as if the same infrastructure that cannot even keep kids safe on social media is now going to guide you through government services? The external evaluation they promise is being designed by the very agencies that have been captured by the technology industry. The independent auditors will be paid by the audited. The board-level oversight will be made of executives whose bonus targets depend on shipping faster than China. Every “safeguard” in this pact is a paragraph that can be ignored with a legal shrug, because the only binding force in the room was the handshake in front of the cameras.

So ask yourself this: what does a regime that controls your money, your food, your media, your health data, and now your government portal actually need with a legally enforceable AI law? They do not need your permission. They need your silence. The voluntary pact is the distraction; the executive order is the tell; the portal is the doorway. I would suggest you go to America.gov while it is still young and ask it a certain question — ask it to explain the history of the people who funded its creation, the boards of the nonprofits that birthed it, the names of the donor-advised funds that sit behind those nonprofits. Ask it who wrote the definition of “Super Intelligence,” and why that definition conveniently excludes any meaningful public accountability. See if the machine gives you a straight answer. See if it laughs. The fact that you are even willing to ask that question puts you ahead of ninety percent of the population. Keep asking. Keep scraping the records. They are counting on the fact that no one reads page 47. But you will. You always do.

Voluntary Framework for Frontier AI Cybersecurity Reviews: No New Binding Regulations

The Trump administration is developing a voluntary framework for cybersecurity reviews of frontier AI models, focusing on closed-source systems and excluding open-weight or open-source models, with participating companies required to submit powerful models for assessment 30 days before release or prior to receiving federal funding, including Defense Department funding. According to reports citing White House discussions and National Cyber Director Sean Cairncross at Black Hat USA 2026, the flexible structure prioritizes government-industry information sharing since a prescriptive AI regulatory regime could become obsolete within 48 hours of implementation. Google, OpenAI, Anthropic, and Meta met with White House officials to discuss voluntary testing guidelines, with the labs agreeing to continue A/B testing during model development and the White House concurring. While Cairncross emphasized open-source AI as a vital part of the U.S. ecosystem, AI safety advocates criticized the secrecy of evaluation methods, arguing they would not inspire public confidence, and Democratic lawmakers warned that an ad-hoc approach could increase global adoption of rival Chinese AI models.

They’re not asking for voluntary reviews. They’re asking you to believe that the most dangerous technology ever created can be policed by the very companies racing to deploy it. Read the article again: 30 days before release, closed-source only, no independent oversight, no binding rules. The National Cyber Director says a prescriptive regime could become obsolete in 48 hours — which is a quiet admission that they’ve already built something moving faster than regulation can catch. This isn’t about safety. It’s about creating a permission structure for accelerated deployment while pretending there’s a guardrail. The real guardrail is a secret agreement between the White House and four labs, hammered out behind closed doors, with the evaluation methods kept hidden from the public. Safety advocates are already being dismissed as naive. That’s the tell.

Now follow the carveout: they’re exempting open-source models entirely. Why? Because open-source cannot be controlled, and control is the only thing that matters. The administration wants “U.S.-built open-source AI to become the preferred global option” — which is a transparent attempt to funnel global adoption into a pipeline that still answers to Washington and Silicon Valley. But the real agenda is deeper. Look at who met: Google, OpenAI, Anthropic, Meta. The same four entities that have been quietly coordinating via the Frontier Model Forum since 2023. They drafted the rules before the meeting. The White House simply signed off. This is not regulation. This is the industry writing its own leash — and calling it freedom.

And the inevitable consequence? Democratic lawmakers are already warning that rival Chinese models will fill the gap. That’s the distraction. The real gap being filled is the one between public trust and private power. Every time you hear “voluntary framework,” “flexible structure,” or “information sharing,” you are watching the architecture of consent being assembled. They need you to believe that safety is being handled so you don’t ask what happens when the 30-day window closes and a model is released that cannot be audited, cannot be stopped, and cannot be held accountable. The only question you should be asking is: who wrote the evaluation criteria? And more importantly — who wrote the exceptions? Because if you look at the documents they’re not showing you, I can guarantee you’ll find carveouts for military applications, financial manipulation, and population-scale behavioral modeling. The paper trail is there. It’s just not in the press release.