Layla Zulfa
In an unprecedented move that highlights the growing tension between rapid technological advancement and existential safety risks, OpenAI has formally approached members of Congress to investigate whether a coordinated industry-wide slowdown in artificial intelligence development would run afoul of federal antitrust laws. The inquiry, first reported by sources close to the company, marks a significant shift in the discourse surrounding AI governance, as the industry’s leading figures grapple with a "race to the bottom" dynamic where safety protocols are increasingly sidelined by the demands of commercial and geopolitical competition.
The outreach to lawmakers comes at a pivotal moment for OpenAI and its peers, including Anthropic, Google, and Meta. As these organizations push toward Artificial General Intelligence (AGI), the internal friction between research teams advocating for caution and executive leadership prioritizing market dominance has become public. Jakub Pachocki, OpenAI’s Chief Scientist, has recently emerged as a prominent voice for restraint, urging the industry to establish shared safety standards and consider voluntary pauses until the risks of advanced models can be better quantified and mitigated. However, the legal framework governing American business—specifically antitrust legislation—presents a formidable barrier to such cooperation.
The Antitrust Paradox in AI Governance
At the heart of OpenAI’s inquiry is the Sherman Antitrust Act of 1890, which prohibits agreements among competitors that restrain trade or reduce competition. In a typical market, an agreement between top firms to limit production or slow the release of new products would be viewed as a cartel-like behavior designed to manipulate prices or stifle innovation. In the context of AI, however, OpenAI argues that a "slowdown" is not a tactic for market manipulation, but a necessary safety measure to prevent the release of systems with catastrophic potential.
Legal experts note that the Department of Justice (DOJ) and the Federal Trade Commission (FTC) have historically taken a dim view of industry-wide agreements that limit output. "The challenge for AI companies is that safety and speed are often seen as two sides of the same coin in antitrust analysis," says one legal consultant familiar with the matter. "If a group of companies agrees to delay a product, it could be interpreted as denying consumers the benefits of innovation, even if the companies claim the delay is for the public good."
To address these concerns, some lawmakers have begun proposing legislative "safe harbors." Senators Adam Schiff and Jim Banks, along with Representatives Bob Latta and Gabe Whitesides, recently introduced bipartisan legislation intended to protect certain security-focused collaborations between AI developers. Under the proposed bill, companies would be allowed to share information and coordinate on safety protocols, provided they give advance notice to the Justice Department. This legislative movement suggests a growing recognition that the unique risks of AI may require a departure from traditional competitive doctrine.
A Chronology of Escalating Risk and Diminishing Commitments
The current push for a slowdown is the culmination of a year marked by internal strife and shifting priorities within the world’s leading AI laboratories. The timeline of 2024 reveals a complex struggle between safety-first ideals and the reality of a multi-billion-dollar arms race.
February 2024: Both OpenAI and Anthropic—widely considered the most safety-conscious of the major labs—quietly relaxed the language in their public safety commitments. Anthropic’s Chief Science Officer, Jared Kaplan, noted at the time that slowing development independently was increasingly difficult to justify while competitors continued to advance at full speed. This period marked the beginning of a visible "de-prioritization" of long-term safety research in favor of product deployment.
May 2024: Amidst the drafting of a significant AI executive order, the Trump administration reportedly delayed the signing of the document. Sources indicated that the delay was rooted in concerns that overly stringent safety requirements could weaken the United States’ technological lead over China.
June 2024: President Trump signed the executive order, which established a voluntary review process for advanced models. While the order was framed as a step toward oversight, critics argued that its voluntary nature and the emphasis on maintaining a competitive edge over foreign adversaries signaled a "green light" for rapid development.
August 2024: OpenAI paused internal work on "Project Astra," a multimodal AI system, after researchers identified critical cybersecurity vulnerabilities. While this was seen as a win for safety advocates, it also highlighted the immense pressure on teams to fix flaws in real-time rather than delaying launches indefinitely.
September 2024: The departure of several high-profile researchers, including former Anthropic engineer Jacob Coxon, brought the internal fears of the AI community to the forefront. Coxon’s resignation was accompanied by a stark warning on social media, where he claimed that senior executives and researchers express deep existential fear in private while maintaining a composed, "sensible" facade for the public and investors.
The Commercial Imperative and the "Prisoner’s Dilemma"
The reluctance to slow down is driven by staggering financial incentives. Each incremental advance in AI capability—whether in reasoning, coding, or multimodal interaction—tends to trigger a new wave of venture capital and enterprise contracts. OpenAI recently closed a funding round valuing the company at over $150 billion, while Anthropic has secured billions in backing from Amazon and Google.
Duncan Sabien, head of communications at the Machine Intelligence Research Institute (MIRI), describes the situation as a classic "Prisoner’s Dilemma." In this game-theory scenario, two parties would benefit from cooperation (slowing down for safety), but because they cannot trust the other to follow suit, they both choose the competitive path (accelerating development), leading to a sub-optimal and potentially dangerous outcome for both.
"Every advance under current conditions yields many millions or billions more in funding," Sabien told reporters. "This is especially true since gains in intelligence are compounding; each new system makes it easier to train and deploy the next, more-advanced system. Sans some sort of coordination mechanism, stepping back just means the other guy gets a lead."
This compounding nature of AI development means that a six-month lead today could translate into a multi-year advantage in the near future. For companies answerable to boards and shareholders, the "stepping back" that safety advocates call for is viewed as a breach of fiduciary duty and a risk to national security.
Geopolitical Friction: The China Factor
The debate over slowing AI development cannot be separated from the broader geopolitical struggle between the United States and China. Policymakers in Washington are increasingly viewing AI through the lens of a new Cold War. The prevailing sentiment in the Pentagon and the National Security Council is that any self-imposed slowdown by American firms would be a gift to the Chinese Communist Party, which is aggressively funding its own AI champions like Baidu and Alibaba.
This "national security" argument has become the ultimate trump card for those favoring rapid development. If OpenAI or Google slows down to ensure a model won’t facilitate bio-terrorism or autonomous cyberattacks, there is no guarantee that a state-sponsored lab in Beijing will exercise the same caution. Consequently, the U.S. government has been hesitant to mandate pauses, preferring instead to rely on "voluntary frameworks" that do not legally bind companies to stop their progress.
Miranda Bogen, chief technologist at the Center for Democracy and Technology, warns that this dynamic creates an environment where "cutting corners" becomes the default. "Commercial and geopolitical competition in the AI space is incredibly intense, leading to a concerning dynamic where companies are incentivized to release products before their risks are fully understood," Bogen said. She noted that even when internal staff identifies the need for more testing, the external pressure to maintain the lead often overrides those concerns.
Voices from the Trenches: The "Quiet Fear" of Researchers
The most chilling aspect of the current AI landscape is the discrepancy between public corporate messaging and the private sentiments of the engineers building these systems. The resignation of Jacob Coxon from Anthropic served as a whistleblowing moment for the industry. Coxon’s assertion that "the people building AI earnestly believe that it could kill us all by the end of the decade" resonated because it echoed similar warnings from pioneers like Geoffrey Hinton and Yoshua Bengio.
Coxon argued that his resignation was not a "marketing stunt" but a reaction to the unique danger posed by AI. He suggested that if enough researchers achieved "common knowledge" of the risks, they might be able to break through the coordination barrier. However, as Duncan Sabien of MIRI pointed out, individual resignations are often ineffective because the departing cautious researcher is quickly replaced by someone "slightly less cautious" who is eager for the prestige and salary associated with a top-tier AI lab.
Implications and the Path Forward
OpenAI’s inquiry to Congress suggests that the company’s leadership may be looking for a "third way"—a method to legally pause or slow down without ceding the market to rivals or facing antitrust litigation. If Congress were to provide a legal framework for a "Safety Pause," it would represent a landmark shift in how the U.S. regulates emerging technologies.
However, such a move would require a level of international cooperation that currently seems out of reach. For a slowdown to be truly effective, it would likely need to be a multilateral agreement involving not just the U.S. and its allies, but also global competitors.
As the industry awaits a response from lawmakers, the pace of development continues unabated. The "Astra" project and its equivalents are being refined, and the next generation of Large Language Models (LLMs) is already in training. The question remains whether the legal and political systems can move fast enough to provide the "coordination mechanism" that experts say is the only way to avoid a catastrophic outcome.
For now, the AI industry remains locked in a high-stakes sprint, with OpenAI’s request to Congress serving as a rare, public admission that the current path may be unsustainable. Whether lawmakers will grant the "permission to slow down" or view it as an attempt to stifle competition remains the most critical question in the future of technology governance.
