The US combines labs' frontier work into one government project

Where things stand (as of Oct 17, 2026)

I found no qualifying publication since the Sept 16, 2026 opening date, and nothing in federal policy is moving toward a combined, government-directed frontier project.

• The President has ruled out nationalization. In his TIME interview (conducted Sept 28, published Oct 1, 2026), Trump was asked whether he would nationalize the top frontier labs and answered "No, no." He named DOJ/FBI as his "guardrail." He was open only to Intel-style equity stakes ("I might"). An equity stake would not count here: funding or ownership alone is not government direction of an integrated project. https://time.com/article/2026/10/01/donald-trump-2026-interview-transcript/
• The federal response so far is coordination and self-regulation. The Sept 29 White House Accord on Super Intelligence is voluntary and "morally binding." The Super Intelligence Force (chaired by DNI Clayton, charter adopted Oct 15) has a 120-day mandate ending around Feb 2027. Its work covers carrying out the Accord, lab incident and breach reporting protocols, critical infrastructure, and enforcement through the FTC and DOJ. Its charter stresses "preventing overregulation and regulatory capture that would stifle innovation and competition." The risk study is led by David Sacks, who opposes even FINRA-style oversight. (TechCrunch Oct 4; Semafor Oct 15; Reuters Oct 16, 2026.)
• The June 2026 executive order rules out mandatory controls. EO 14409 sets up only a voluntary 30-day pre-release access framework and expressly disclaims any licensing or preclearance. https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/
• Industry proposals are about coordination, not merging. Amodei's "pacing" plan calls for embedded evaluators, coordination among democratic-country labs, and global coordination. Hassabis proposes a FINRA-like standards body. OpenAI wants international standards. Google, OpenAI and Anthropic are building their own industry body (SAFA). None of these proposes integrating labs' development under the government.
• Congress is not proposing it either. Democratic proposals cover standards agencies, kill switches, liability, and the Sanders/Casar ban plus a new Department of AI.
• The Genesis Mission is not a match. It remains an AI-for-science effort (grants, $2.4B in compute pledges announced Oct 8), so it does not meet the question's definition of frontier AI work.

Signals pointing the other way. The government is intervening more and more:

• In June, a Commerce export-control order shut down Anthropic's top models for 19 days.
• In February, the Pentagon designated Anthropic a supply-chain risk.
• Vance and Hegseth have floated using the Defense Production Act.
• The national-security establishment now leads AI policy, under a "whoever wins AI wins" framing.
• Labs say recursive self-improvement is beginning, and China is reportedly only months behind.

This is the setting in which a government-run AGI "Project" (as Aschenbrenner and the AI 2027 scenario describe) becomes conceivable.

Reference class and base rate

• History. In peacetime, the US has essentially never merged rival firms' core R&D under government direction. Wartime mobilizations did so: synthetic rubber, aviation fuel, the radiation lab at MIT, and the Manhattan Project.
• Expert forecasting. An IAPS workshop of professional forecasters and policy experts (2025) put the chance at 34% that a US government-led project builds the first system that speeds up AI research tenfold, conditional on that capability arriving by 2035. The most likely form was a government-led consortium (14%), and participants put the chance of such a project involving more than one company below 50%. https://www.iaps.ai/research/us-government-role-in-advanced-ai-development
• Forecasting sites. Metaculus has "US government control of any US AI company or project before 2029" at about 17–20%. That is a broader event; most of those scenarios involve a single lab. Manifold markets on this topic (28% for an AGI "Manhattan Project" before 2029; 6.8% for "much AI research nationalized by 2027" with 2.5 months left) look inflated by long-shot bias.

Scenario estimate

• Fast progress (~45%): superhuman automated AI research by 2028. The chance of a qualifying, officially announced multi-lab consolidation is about 8–10%.
• Slower transformative progress (~35%): about 4%.
• Plateau (~20%): about 0.5%.

This gives roughly 5–7%. The strict criteria pull it down further: the announcement must name the developers, describe integration with a shared technical plan, and give a federal body authority over that plan. Parallel contracts, consortia, equity stakes and single-lab takeovers do not count, and a classified effort might never be described in enough detail.

Hazard series (computed)

Per-quarter hazards:

• Q4 2026 (partial quarter): 0.12%
• 2027: rising from 0.22% to 0.36%, as the task force report lands and capabilities climb
• 2028: 0.36–0.40%, Trump's final year with a possible crisis or race-driven push
• 2029: 0.40–0.46%, a new administration's first year
• 2030–31: tapering from 0.36% to 0.29%

Cumulative probability (one minus the running product of 1 − hazard): about 0.1% by end-2026, 1.3% by end-2027, 2.8% by end-2028, 4.5% by end-2029, and about 7.0% by end-2031.

Strongest case against my estimate

This world looks a lot like AI 2027: rogue-agent incidents, labs pausing training, and early recursive self-improvement. If superintelligence looks imminent and China close, a president who likes deals and grand projects, with a DNI as AI czar, could announce something like an "American Super Intelligence Project" merging labs' compute and teams. That could push the five-year figure into the low teens. I do not go that high because every actual lever now in use (accords, voluntary review, export controls, FTC probes, equity talk) points to oversight of competing labs, not integration. Labs worth trillions of dollars and heading toward IPOs, antitrust law, and the administration's pro-competition stance are all large obstacles.

Question. By the last day of each quarter, has the US government announced that the frontier AI work of two or more covered developers is being combined into a single project under government direction?

Definitions.

• Frontier AI work: research, development, or training aimed at producing or materially advancing a developer’s most capable general-purpose AI models or systems. Providing access to existing models, applying them to scientific or military tasks, and developing only narrow application-specific tools do not by themselves qualify.
• Combined into a single project: at least two distinct covered developers contribute frontier AI work to one integrated development effort with a shared technical development plan and common model or system deliverable. The announcement must identify the developers and describe the integration of their research teams, training efforts, or model-development resources. Separate projects within an umbrella initiative do not count. Combining all of either developer’s work, merging the companies, and pooling model weights are not required.
• Under government direction: a US federal executive department, agency, or government-appointed project leadership has authority to set or approve the integrated project’s technical development plan and to direct participating developers’ project work or allocation of development resources. Funding, procurement requirements, regulatory oversight, or advisory participation alone are insufficient. Voluntary participation counts if this project-level authority exists.
• Announce: officially state that the combination has been authorized or agreed and will proceed, or is already occurring. A proposal, request for information, invitation to collaborate, negotiation, or aspiration does not count.

Resolves YES if a qualifying federal announcement establishes all the above elements. They may be established by the announcement together with its expressly incorporated public agreements or governance documents. An authorized project may qualify before operations begin; later cancellation does not reverse YES.

Exclusions (resolve NO).

• Parallel contracts with multiple developers, shared computing facilities, common datasets, model marketplaces, safety-evaluation collaborations, or government access to multiple models without integrated frontier development.
• A consortium or national initiative whose members retain separate development programs and whose government leadership merely coordinates participation.
• A merger or private joint venture without the specified federal direction.
• The Department of Energy’s December 18, 2025 Genesis Mission collaboration announcement and the Genesis Mission Consortium description alone do not establish an integrated frontier-development project involving two covered developers. A subsequent announcement under that initiative counts only if it independently meets these criteria.
• Resolve NO for a horizon if no qualifying announcement has been published by its end.

Resolving source. Official announcements, executive orders, published agreements, or official transcripts issued by the White House or a US federal executive department or agency, including Defense, Energy, or Commerce, on its official website or in the Federal Register. This is a narrower source requirement than Public confirmation (A2): developer statements and third-party reporting cannot independently resolve the question. Apply the Resolution window and Resolution date (A3) conventions.