US task force assigns work on voluntary superintelligence accord


Task-force chair Jay Clayton says a risk study and three committees will help put voluntary company safety commitments into practice.
The US task force on superintelligence (SI), or artificial intelligence, set up a risk study and three committees at its first meeting in Washington on 15 October. National intelligence director Jay Clayton described the assignments in a 16 October CNBC interview, Reuters reported.
The work follows the White House Accord on Superintelligence, which two dozen technology companies agreed to on 29 September. The commitments include independent safety audits but impose no new legal requirements, Ars Technica reported on 30 September. The new assignments include developing ways for laboratories to warn the government about security incidents; they are not binding notification requirements.
Clayton said the meeting established a path for implementing the accords and maintaining dialogue with laboratories developing the most advanced models. David Sachs will lead a risk study. Reuters reported these committee assignments from Clayton's CNBC account:
| Committee | Lead |
|---|---|
| Reporting and response | Emil Michael, US Defense Department technology lead |
| Critical infrastructure | Scott Kupor, Office of Personnel Management director |
| Law enforcement and responsibility |

DeepSeek says its new 552-billion-parameter model's cache needs much less memory than the previous generation's and beats V4-Pro in tests. The company says V4-Pro requests are due to route to it from September 14 and has changed its API prices. The test claims come from DeepSeek itself.
A White House account of the meeting, released on 15 October and reported by Reuters, assigns Michael to review and develop ways for laboratories to provide early warnings and systematic notifications of breaches, hacks, jailbreaks and other problems. A jailbreak is an attempt to bypass a model's safeguards.
The White House also assigned Clayton and national cyber director Sean Cairncross to lead a working group on implementing the accord. Clayton said workforce and economic impacts would be reported to the US Department of Labor, while enforcement against wrongdoing involving the technology would run through the Federal Trade Commission or the US Justice Department.
Trump announced Clayton's appointment to lead the Super Intelligence Force on 4 October. Fortune's account of Trump's announcement said the group would coordinate federal work, with the stated aims of maintaining US leadership and protecting Americans' interests. It would report to Trump and his chief of staff, Susie Wiles, and consult consumers, public-interest groups, religious organizations, critical infrastructure providers and technology companies.
TechCrunch reported on 4 October, citing The Wall Street Journal, that the task force's charter called for threat-response plans while seeking to prevent excessive regulation and regulatory capture from stifling innovation and competition. Regulatory capture means regulators serving industry interests rather than the public.
The Register reported on 30 September that Meta, Google, OpenAI, Anthropic, Nvidia and xAI were among the accord's signers, and that Trump posted the document on his Truth Social account.
The accord lists four "layers of controls and audits" that the White House believes signers should adopt "in addition to any other precautions", The Register reported:
The companies say they "believe that implementing these controls and audits is critical to ensuring a safe future for everyone" and that each is "committed to doing this."
Analysis
The Register noted that the accord specifies neither the required rigor nor the frequency of checks and contains no enforcement mechanism. The document says companies "will meet regularly to establish standards and best practices to improve the safety of their systems." The Register interpreted this as companies getting to make their own rules and called the accord weak and vague.
On legal force, the document says "it may make sense to codify these steps into laws or regulations" without committing to do so. Clayton's account of enforcement through federal agencies does not turn the voluntary accord into a legal requirement.
Trump's 29 September executive order requires federal agencies to use "Super Intelligence" and "SI" instead of "Artificial Intelligence" and "AI" in official documents and communications, The Register reported.
The order gave Trump's Assistant for Science and Technology 60 days to work with other agencies on a definition. Its stated starting point was that the most advanced systems "do much more than imitate or automate discrete aspects of human intelligence" and "increasingly amplify human ingenuity and unlock new forms of creativity."
The Register linked the accord and the naming order, arguing that they reflected Trump's belief that the industry did not need strong regulation except over the technology's name.
The accord leaves companies to establish standards and does not specify how they must demonstrate compliance. Ars Technica reported on 30 September that the choice of outside auditors and the way their tests would work had not been settled.
The naming order also requires a federal definition within its 60-day window, to guide the terminology agencies have been ordered to use. The task force's assigned work now includes the risk study, accord implementation and development of laboratory incident-warning protocols, according to Clayton's account and the White House description reported by Reuters.



4 min

The security firm Mindgard says it got past the safety limits on two of Moonshot's Kimi models. It has not tested whether the answers would work. Moonshot says it is reviewing the findings and that its own tests show the models usually refuse such requests.
4 min