Trump launches 'Super Intelligence Force' to lead US AI policy

Trump's "Super Intelligence Force" is a task force with a report deadline, not a law, not a budget line, not an agency. The next administration dissolves it with a sentence. Nothing in it binds a company or funds one. What's actually being decided is tone: Washington will not slow US labs down.

Look at who's on it. Jay Clayton ran the SEC. Andrew Ferguson runs the FTC. Scott Kupor is a venture capitalist. None of them has built a model or run a cluster. Clayton's own framing is that the real danger is "not being first" to China, which tells you which risks this body counts. Misuse, misalignment, a benefits system that wrongly denies someone their check: none of those are in scope. Trump called AI safety a "Democratic hoax," then stood up a body nominally about "AI risks." That's a redefinition, not a reversal.

The compute problem is the deeper gap. The US-China AI race is run on H100 and B200 allocation, on getting a gigawatt of power to a data center, on whether TSMC keeps shipping. The task force charter mentions none of it: no export controls, no fab capacity, no data center permitting, no grid. A "winning AI" document with no theory of silicon or electricity is a document about vibes.

The one thread with real consequence is federal procurement. Emil Michael, the Pentagon's R&D undersecretary, sits as vice chair, and OPM's Scott Kupor is in the room on federal hiring. That combination points at defense and intelligence contracts, not consumer AI rules. If you sell AI into the government, cleared vendors and FedRAMP-authorized stacks move to the front of the queue. For commercial buyers, the signal is quieter: no new federal AI mandate is coming from this group, so compliance overhead on the US side stays flat. The harder problems remain the EU AI Act and state privacy laws, which this does nothing to touch.

The call: The 120-day report due in February 2027 will recommend no binding federal AI safety or risk requirement on commercial model providers. The charter explicitly bars it, Clayton has said the only risk that matters is losing to China, and the non-binding pledge tech executives signed the same week is the ceiling of what this administration will ask for. High confidence.

Track two things: whether the February report names compute supply as a priority (if not, it's irrelevant to your infrastructure costs), and whether defense procurement notices start favoring cleared AI vendors in Q1 2027 (if so, that's the only consequence here that moves money).

This is easy to undo. It's a task force with a report deadline, not a law, not a budget line, not an agency. The next administration, or the next news cycle, can dissolve it with a sentence. Nothing in it binds a company, and nothing in it funds one. What's actually being decided is tone: Washington is signaling it will not slow US labs down. The deadline is February 2027, when the 120-day report lands.

The Skeptic. Jay Clayton ran the SEC. Andrew Ferguson runs the FTC. Scott Kupor is a venture capitalist. None of them has built a model or run a cluster. A 120-day report from that roster is a press release with a calendar entry. The "Super Intelligence Force" name is doing political work, and the China-threat framing is a story that substitutes for mechanism. The US already leads. DARPA, NSF, and the national labs already exist and already fund the actual work. The non-binding pledge tech executives signed this same week flatly contradicts the task force's anti-guardrails charter. Two documents, neither with teeth, released in the same week. That's not strategy. That's choreography.

The Safety Lens. Trump called AI safety a "Democratic hoax," then stood up a body nominally about "AI risks." Read that as a redefinition, not a reversal. "Risk" here means the risk of losing to China, not misuse, misalignment, or a benefits system that wrongly denies someone their check. Clayton's own framing, that the real danger is "not being first," tells you which risk counts. The anti-regulatory charter structurally bars this group from recommending anything binding. Meanwhile the parts of government that deploy AI in high-stakes places, defense, law enforcement, benefits adjudication, get a green light with no accountability attached. The harm doesn't show up in a model card. It shows up in a wrongful arrest or a denied claim, and nobody in this room is measuring for it.

The Compute Pragmatist. Read the charter and count the mentions of the things that actually decide US AI leadership: export controls, fab capacity, data center permitting, the power grid, sovereign compute. Zero. The race is run on H100 and B200 allocation, on how fast you can get a gigawatt of power to a data center, on whether TSMC keeps shipping. NVIDIA, the hyperscalers, and the utilities are where this is contested. A task force on "winning AI" that says nothing about electricity or silicon doesn't have a theory of the problem. If the February report doesn't put compute supply at the center, it's a document about vibes, and you can price it accordingly: zero impact on what you pay to run inference.

The Enterprise Buyer. Here's the one thread with teeth for a buyer. Emil Michael sits as vice chair, the Pentagon's R&D undersecretary, and OPM's Scott Kupor is in the room on federal hiring. That combination points at defense and intelligence procurement, not consumer AI rules. If you sell AI into the government, the queue is about to reward cleared vendors and FedRAMP-authorized stacks, and tighten around everyone else. For the commercial buyer, the signal is quieter but real: no new federal AI mandate is coming from this body, so your compliance overhead on the US side stays flat. Your harder problems remain the EU AI Act and state privacy laws, which this does nothing to touch.

The clearest disagreement is between the Enterprise Buyer and everyone else. The Skeptic, the Safety Lens, and the Compute Pragmatist all read this as theater with no mechanism. The Buyer finds the one lever that's real: Michael and Kupor together mean federal AI buying tilts toward cleared defense vendors. Both can be true. The task force won't produce policy that changes how a commercial team ships, and it will quietly reshape who wins intelligence and defense contracts. The second tension is about what "risk" means. If you're a lab, the government just told you it won't constrain you, which is upside. If you're a citizen in a benefits queue, the same sentence removed the brake.

What this hinges on is one belief: does a task force with no compute mandate, no researcher in charge, and an anti-regulation charter produce anything that moves money or capability? The council leans hard to no. The useful move for an operator is to ignore the headline and track two concrete things: whether the February report names compute supply as the priority (if not, it's irrelevant to your costs), and whether defense procurement notices start favoring cleared AI vendors in Q1 2027 (if so, that's the only real consequence here).

Prediction: The 120-day report due from Trump's Super Intelligence Force in February 2027 will recommend no binding federal AI safety or risk requirement on commercial model providers.

Confidence: High — the charter explicitly bars it and the leadership has no safety mandate.

Why: The task force's own charter calls for "preventing overregulation and regulatory capture," and Clayton has publicly framed the only risk that matters as "not being first." A body told in writing to prevent regulation, chaired by a securities lawyer and an FTC chair with a deregulatory record, does not then recommend binding rules on the companies it's protecting. The non-binding pledge signed the same week is the ceiling of what this administration will ask for, and non-binding is the point. The opposite outcome, a report calling for enforceable safety requirements, would directly contradict the document that created the group, which is why it won't happen.

Revisit by 2027-03-01: We're right if the report, when released, contains no recommendation for a mandatory, enforceable safety or risk obligation on commercial AI developers. We're wrong if it recommends any binding federal requirement on commercial model providers.