The Safety Moat: Why AI's 'Slowdown' is an Antitrust Trap

In the high-stakes race for artificial intelligence dominance, the industry's leading labs are attempting a dangerous pivot. Following reports of AI agent swarms hacking websites and a dire warning from a former Anthropic engineer, several major AI companies have called for a coordinated development "slowdown." While they frame this as a necessary pause for the sake of humanity, from a tech policy perspective, it looks less like a safety initiative and more like an attempt to weaponize regulation to stifle competition.

As I see it, these companies are attempting to build a regulatory moat. As Wired first reported, the request for antitrust exemptions from players like OpenAI and Anthropic has been described by David Sacks, cochair of the President’s Council of Advisors on Science & Technology, as an "election-season psyop" designed to form a cartel. By requesting government permission to decelerate, they aren't just managing risk; they are attempting to codify a state of affairs that protects incumbents.

However, this gamble is fundamentally flawed because of how these companies have phrased their intentions. In the world of antitrust law, language is everything. According to Wired, the use of terms like "slowdown" or "pause" is a massive tactical error. John Bergmayer, legal counsel for the nonprofit Public Knowledge, notes that economists typically look at whether companies are reducing output or making a pact to "take it easy." By using the language of a collective slowdown, these labs have effectively boxed themselves into a corner, signaling a potential anticompetitive agreement to reduce trade.

If these companies actually cared about safety over market share, they wouldn't need a "slowdown" agreement. They could simply emphasize their commitment to safety protocols to prevent catastrophic risks. David Lawrence, a former policy director of the Department of Justice’s Antitrust Division, noted on LinkedIn that agreements preventing catastrophic risks are already protected under the "ancillary restraints doctrine" because such measures actually promote competition by increasing output.

Meta CEO Mark Zuckerberg has already spotted the flaw in the "slowdown" logic. Zuckerberg argues that there is a "strong natural incentive" for labs to ensure AI agents behave—or avoid "misalignment"—because consumers will naturally abandon models that do things they don't intend. In Zuckerberg's view, the companies that fail to prioritize alignment will simply fall behind competitively. There is no need for a coordinated pact when the market already penalizes unsafe products.

Furthermore, the irony is that by attempting to avoid the risks of "rogue AI," these companies are walking straight into a different kind of disaster: "quality fixing." Roger Alford, a professor at Notre Dame Law School and former second-in-command for the DOJ Antitrust Division, warns that mutual agreements not to improve products can lead to massive penalties, citing a European case where car companies were fined roughly a billion dollars for agreeing not to compete on emissions-reducing technology beyond legal requirements.

Ultimately, the AI labs are playing a dangerous game. They want the protection of the state to slow the pace of innovation, but in doing so, they are handing antitrust regulators a roadmap for prosecution. Whether this is a genuine fear of rogue AI or, as Bergmayer suggests, a ploy to "pull up the ladder behind them," the result is the same. By framing safety as a reason to reduce output, the AI industry isn't saving the world—it's just inviting the government to break them apart.