Asger Risom
The new White House AI safety agreement is a masterclass in corporate lobbying. By accepting a “morally binding” pact based entirely on voluntary self-policing, Washington has formally surrendered to Silicon Valley, trading the safety of the digital economy for a polite corporate handshake.
When the federal government sits down with the architects of the most disruptive technology in human history, you expect them to draft legislation. Instead, Washington just accepted a pinky promise.
The new AI safety agreement announced by the White House this week is a masterclass in corporate lobbying disguised as public policy. Major tech executives—including Meta’s Mark Zuckerberg, Nvidia’s Jensen Huang, and Anthropic’s Dario Amodei—secured a framework that the Trump administration described as entirely “morally binding.” The deal relies on “tremendous self-policing,” involves zero actual government regulators, and mandates no public disclosures of internal audits.
A “constitution” with no laws
For the past year, tech executives have paraded through congressional hearings, warning of existential risks and positioning themselves as the responsible adults in the room. But behind closed doors, Silicon Valley worked to ensure that when the actual rules were written, they wouldn’t have any teeth.
A “morally binding” agreement is not regulation. It is a legal forcefield. By agreeing to four voluntary steps—like establishing internal corporate board committees and running private safety checks—tech giants have successfully neutralized the threat of federal oversight. President Trump went so far as to praise the voluntary pact as “almost like a constitution.” Meanwhile, the administration used the summit to officially rebrand artificial intelligence as “super intelligence” (SI)—a move that feeds directly into the industry’s sci-fi marketing hype while explicitly promising not to enforce strict guardrails.
In exchange for these vague assurances, tech companies get to slap a government-approved safety seal on their products while maintaining absolute secrecy over their training data, safety failures, and error rates.
Outsourcing public safety
The mechanics of this agreement are fundamentally broken. If a pharmaceutical company releases a toxic drug, or an automaker builds a defective brake line, federal agencies have the statutory authority to audit the failure, demand the data, and issue penalties.
Under this new AI pact, if a language model generates non-consensual deepfakes at scale, automates mass phishing attacks, or compromises enterprise security, the government’s only recourse is to hope the offending company’s board of directors handles it internally. We have entirely outsourced the safety of the digital economy to the very billionaires who are profiting from its disruption.
Washington didn’t regulate artificial intelligence this week. They formally abdicated their responsibility to do so, trading the safety of the American public for a polite corporate handshake.
