U.S. AI Regulation Scramble Exposes Missing Foundation

The United States government's recent negotiations with OpenAI and Anthropic over their most powerful AI models represent a regulatory process that policy experts argue should never have been necessary in its current form.

Both companies secured government approval before releasing their latest systems—a milestone that would have seemed impossible months earlier. Yet behind the public statements of successful collaboration lies a makeshift process characterized by export control threats, licensing requirements, and coordination challenges across multiple agencies.

Why it matters

The ad-hoc nature of AI oversight in the U.S. reflects deeper structural failures: inadequate government technical capacity, underfunded regulatory agencies, and the absence of congressional legislation. These gaps force both companies and regulators into reactive, case-by-case negotiations that create uncertainty and may not effectively address safety concerns.

The Jailbreaking Problem

The regulatory scramble became visible when Amazon flagged vulnerabilities in Anthropic's systems that allowed users to bypass safety guardrails—a practice known as jailbreaking. This discovery ultimately triggered export controls on the company.

The Biden administration had required companies to share safety testing results, including jailbreaking vulnerabilities. President Trump eliminated those reporting requirements as part of his deregulatory agenda. When safety concerns with Anthropic emerged, no shared framework existed to assess the severity of such vulnerabilities.

According to a source familiar with the situation who spoke with Axios, a standardized assessment framework might have prevented the export control situation entirely.

Systemic Deficiencies

Former Biden tech official Asad Ramzanali noted that other countries have implemented privacy regulations, updated antitrust laws, and established transparency requirements. "We didn't do any of it," he said, leaving the U.S. without a foundation for AI rules.

The Cato Institute's Kevin Frazier pointed to the government's failure to recruit technical talent—fewer than 1 percent of AI Ph.D. holders enter public service. Key agencies including the Cybersecurity and Infrastructure Security Agency and the Center for AI Standards and Innovation remain underfunded. CAISI operates on a $15 million budget despite needing $84 million annually to execute Trump's AI action plan, according to the Institute for Progress.

Congress has not passed comprehensive AI safety legislation despite years of bipartisan efforts. "Right now, there is far too much confusion with the White House's AI vetting process," Rep. Josh Gottheimer (D-N.J.) said.

Voluntary But Mandatory

The Trump administration describes its cybersecurity executive order provisions as voluntary and states it has not invoked the Defense Production Act. Yet the Anthropic case demonstrates that companies face significant pressure to cooperate. OpenAI CEO Sam Altman acknowledged this reality: "You really want to be confident in your safety claims because otherwise the world is going to get uncomfortable very fast."

The White House maintains it neither approved nor disapproved OpenAI's model release, with spokesperson Liz Huston stating that "top technology leaders in the country are fully lining up behind the President's commonsense approach."

Industry and the administration continue developing the voluntary framework required by the June AI executive order, due August 1.

These details were first reported by Axios.