According to WIRED, quoting sources close to OpenAI, OpenAI recently asked U.S. congressmen whether it would be possible to violate U.S. antitrust laws if several AI companies collaborated to slow down their development pace. This move comes at a time when industry security debates are intensifying, and it also reflects the tension between competition and risk among leading model companies.
Antitrust issues become a focal point of discussion
OpenAI Chief Scientist Jakub Pachocki Previously called for the industry to consider voluntarily slowing down the pace of advancement until developers can more fully demonstrate the security of the models. However, the reality is that even if internal researchers at a single company are concerned about the risks, it is difficult for them to slow down on their own.
According to the current logic of U.S. competition law, if multiple companies reach a consensus to limit development or distribution, it may be considered as weakening market competition, thus posing legal risks. The report mentions that OpenAI has been communicating with members of Congress regarding this issue in recent weeks.
U.S. Senators Adam Schiff and Jim Banks have proposed a bill to provide legal protection for certain cooperative efforts aimed at security purposes, provided that such arrangements must be notified in advance to the Department of Justice. In other words, discussions at the congressional level have already begun regarding whether AI companies can have more explicit coordination on security matters.
Security concerns coexist with financing pressures
Many interviewees believe that what truly hinders the slowdown in development is not just legal issues, but also financial constraints and competitive pressures. Miranda Bogen, the chief technical expert at the Democracy and Technology Center, stated that the commercial and geopolitical competition in the AI sector is very fierce, which makes it easier for companies to launch products before the risks have been fully understood.
Duncan Sabien, the person in charge of communication at the Institute of Machine Intelligence, also stated that in the current environment, every improvement in model capabilities could lead to millions or even billions of dollars in financing, further expanding the influence of developers. For companies like OpenAI and Anthropic, if there is no coordination mechanism in place, one party that slows down may quickly fall behind.
He also pointed out that the improvement in model capabilities has a cumulative effect. New systems often help to train and deploy the next generation of even stronger systems, which makes it more difficult to justify “pausing for observation” from a commercial standpoint.
Industry debates continue to intensify.
The report mentioned that OpenAI suspended some work on its internal Astra project in August due to concerns about network security, as the relevant security measures were insufficient. However, both OpenAI and Anthropic relaxed their security commitments in February of this year. The then Chief Scientific Officer of Anthropic, Jared Kaplan, argued that unilaterally slowing down development was of limited significance when competitors continued to make progress.
This week, the debate has heated up again. The former Anthropic engineer Jacob Coxon publicly announced his resignation, stating that continuing to push forward with the development of AI poses serious risks to humanity. He stated on the social platform X that many people involved in the development of AI privately have genuine concerns about these risks.
However, there are also views that the departure of individual researchers may not necessarily change the pace of the industry. If a broader consensus cannot be reached, those who leave will soon be replaced by others. The core of the current debate is no longer just whether to slow down or not, but whether leading companies can establish some kind of joint mechanism for deceleration within the limits permitted by law.
