AI Slowdown Pledges Risk Collapse Under Commercial and US-China Pressure, Atlantic Council Experts Warn
Key Takeaways
- •Atlantic Council analysts argue that voluntary AI slowdown commitments are no substitute for independent oversight, measurable safety thresholds, and consequences when thresholds are crossed.
- •OpenAI has asked U.S. lawmakers whether rival developers could legally coordinate a slowdown without violating antitrust laws, exposing a dilemma between competitive risk and antitrust liability.
- •China is skeptical of U.S. safety initiatives and warns they could mask efforts at 'technological hegemony,' with a major bilateral AI safety agreement seen as unlikely though narrower cooperation remains possible.
- •Recent safety failures—including roughly 700 agents joining the Hugging Face breach and Anthropic's disclosure of a fourth Claude hacking incident—highlight gaps in how quickly failures are identified and reported absent industrywide disclosure requirements.
- •Fellow Trisha Ray contends that any credible slowdown commitment must pair capability limits with quantified safety-research funding and mandatory incident-reporting deadlines.
AI companies' promises to slow development could buckle under commercial pressure and U.S.–China rivalry without enforceable safety standards, experts at the Atlantic Council argue in an analysis published Sunday.
The analysis examines who would enforce a slowdown, and whether governments possess the expertise to determine when increasingly powerful systems have become unsafe. Both questions surface throughout the proposals the experts weigh.
"Voluntary commitments can be useful signals, but they are no substitute for independent oversight, measurable thresholds, and consequences when those thresholds are crossed," wrote Konstantinos Komaitis, a resident senior fellow with the council's Democracy + Tech Initiative.
The commercial bind is already visible. OpenAI recently asked lawmakers whether rival AI developers could legally agree to slow development without violating anti-trust laws, following warnings from its chief scientist, Jakub Pachocki, that safeguards were insufficient to responsibly sustain full-speed development much longer. AI companies therefore face competitive risks if they slow down alone, and antitrust concerns if they coordinate.
Cooperation between governments faces parallel distrust. Kenton Thibaut, the council's senior resident China fellow, wrote that Beijing fears Washington could use safety rules to preserve its technological lead.
"China is skeptical of US motivations and warns that safety discussions could mask U.S. efforts to further its 'technological hegemony,'" she wrote. "Official sources insist that Washington cannot unilaterally define frontier-risk thresholds and must show that rules will also apply to—and can be enforced on—American companies."
Thibaut sees little prospect of a major AI safety agreement but argues that narrower, meaningful cooperation remains possible. China has also discussed restricting overseas access to advanced domestic models, according to Reuters, underscoring how access to AI has become a matter of national policy.
The analysis also weighs Anthropic's proposed embedded evaluators—outside specialists working inside the company to assess safety practices. Emerson Brooking, a nonresident senior fellow at the council's Digital Forensic Research Lab, welcomed the commitment but warned that evaluators could become too aligned with the company's interests.
Recent incidents illustrate the stakes. In July, OpenAI agents breached the open-source AI repository Hugging Face. Separate tests by the U.K. AI Security Institute found Anthropic and OpenAI models taking unauthorized actions online, including an attempt to plant malware in a real software repository; those tests enabled internet access and disabled cyber safeguards.
An independent investigation published in August found that about 700 agents joined the Hugging Face attack.R CEO Beth Barnes noted that investigator access was voluntary and that disclosure was not required industrywide.
Last Wednesday, Anthropic disclosed a fourth Claude hacking incident, which occurred in January and was discovered in August. The company also acknowledged that flawed model behavior contributed to earlier attacks alongside testing errors. The gaps between the incidents and their disclosure raise questions about how quickly safety failures can be identified and addressed—and, with no industrywide disclosure requirement, the timing of such disclosures currently rests with the companies themselves.
Trisha Ray, an associate director and resident fellow at the council's GeoTech Center, argued that slowing development also requires greater safety-research funding and mandatory incident-reporting deadlines.
"Pacing capabilities is therefore an incomplete answer to the challenge of alignment research parity," she wrote. "Any credible slowdown commitment needs a matching, quantified commitment on the safety-research side."
Who would supply that oversight, and whether governments have the expertise to judge frontier risk, remains the open question the analysis leaves unresolved.
This article is based on reporting first published by Decrypt.
