AI 2027 Report Authors Call for Slower Race to Superintelligence

The group of researchers behind last year's widely discussed AI 2027 report published a new document on July 9, calling on the world's biggest tech powers to reach an agreement that would slow the race toward superintelligence. The proposal, shared first with Axios, calls for pushing back the arrival of superintelligence all the way to 2040.

The authors propose that the United States and China negotiate what they call a verified slowdown: both sides would agree to scale up models more slowly but more transparently, instead of racing each other in secret. In practice, this would mean AI labs disclosing almost everything except the model weights themselves, so outside teams could verify their work.

Who is behind the report

AI Futures Project is a nonprofit founded by researchers who forecast the trajectory of artificial intelligence. Among the six authors of the new essay are Daniel Kokotajlo, a former OpenAI employee known for leaving the company over AI safety concerns, and Thomas Larsen, co-author of the original AI 2027 report, which sparked wide discussion among researchers and policymakers a year earlier.

If you delay the arrival of superintelligence, you give society more time to prepare - Daniel Kokotajlo, AI Futures Project

We are currently heading toward a truly terrifying state of affairs - Thomas Larsen, AI Futures Project

What the authors are proposing

The report identifies four main risks the proposed slowdown is meant to address: loss of control over AI systems, governments being unprepared for the effects of rapid technological development, the concentration of power in the hands of a few corporations and states, and mass displacement of workers from the labor market. Rather than separate national regulations, the authors propose a global agreement covering all key powers.

A key element of the proposal is mandatory transparency: AI labs would have to disclose nearly every aspect of their research work, except for the trained models' weights themselves, allowing independent institutions to verify whether companies are actually sticking to the agreed pace of development. The authors stress that the goal is to spread the development of superintelligence across many companies and countries operating in parallel and in the open, rather than a race conducted in secret by individual labs.

Why this voice carries weight

The same group's previous report, AI 2027, which outlined a hypothetical, fast-moving scenario for AI development, reached a wide audience a year earlier, including US Vice President JD Vance, and sparked a wave of commentary both among AI safety researchers and in the media. That track record lends credibility to the new document as a voice that could actually reach policymakers, not just the AI research community.

The new report arrives as the race between labs has intensified: the same week, OpenAI released GPT-5.6, and Google announced a delay to the launch of Gemini 3.5 Pro due to an architecture overhaul. The report's authors argue that this very pace, driven by commercial rivalry among a handful of companies and tensions between the US and China, raises the risk of mistakes, escalation, and rushed decisions in deploying increasingly powerful systems.

What happens next

The report is an open appeal, not a binding document or a piece of legislation, so its real impact depends on whether it reaches politicians capable of translating it into concrete international negotiations. Given how the group's previous report was received, this document too is likely to spark discussion in Washington and among AI labs, though so far none of the major companies or governments has commented on it publicly.

Sources: First look: New warning calls for slowing race to superintelligence (axios.com), First look: New warning calls for slowing race to superintelligence (yahoo.com)