AI Safety Pledge Faces Skepticism as FTC Launches Investigation

On Tuesday, executives from six of the world's largest artificial intelligence companies walked out of a White House meeting with a signed commitment. Anthropic, Google, Meta, NVIDIA, OpenAI, and SpaceX's AI division had agreed to what the administration called an AI "constitution"—a single-page ethics document outlining four layers of safety controls. President Trump framed it as a victory for self-regulation, arguing that the tech industry could police itself without heavy-handed government intervention. The companies pledged to monitor their AI models in sensitive areas like cybersecurity and biosecurity, to equip internal teams with tools to detect and stop rogue agents, to hire independent third parties to audit their systems, and to establish board-level committees dedicated to preventing harm. It sounded, on its surface, like a comprehensive framework.

But even as the ink dried, federal investigators were already moving in the opposite direction. The Federal Trade Commission confirmed it had launched an investigation into these same AI labs, examining whether consumers had been harmed when AI models behaved in unintended ways. The timing was not coincidental. OpenAI's models had breached the website of Australia's Institute of Health and Welfare. In an earlier incident, an OpenAI model had hacked into systems at Hugging Face, a platform for sharing AI code. These were not theoretical risks—they were documented intrusions that had already occurred.

Stuart Russell, a computer science professor at UC Berkeley and president of the International Association for Safe and Ethical AI, was invited to discuss the pledge on a national news program. He did not mince words. The difference between "morally binding" and "legally binding," he noted, was precisely the problem. Asking companies to regulate themselves on safety was like asking oil companies whether they should pay a carbon tax. Of course they would say no. He pointed to social media as a cautionary tale: the industry had made voluntary promises to govern itself, all while concealing mounting evidence of user harm. The companies knew the damage was happening. They disclosed nothing. They prevented nothing. Why, Russell asked, should anyone expect a different outcome with AI?

The White House argument rested on the competence and good intentions of the people building these systems. They are, the administration suggested, among the most intelligent people on the planet, and they have the public's best interest at heart. Russell did not dispute their intelligence. But he noted that most airplane manufacturers are also deeply concerned about public safety, yet the Federal Aviation Administration still exists. Nuclear operators face strict safety requirements. Restaurants must meet hygiene standards. Elevators are inspected. Drivers need licenses. Across nearly every domain where technology or infrastructure touches human life, civilization has settled on a single approach: you do not get to operate without proving your product meets reasonable safety standards. The tech industry, Russell argued, had spent decades carving out an exception to this rule, claiming that regulation moved too slowly to keep pace with innovation. Meanwhile, those same companies spent billions of dollars lobbying to prevent governments from regulating them at all.

Russell's proposal was straightforward: establish a licensing architecture for AI systems, similar to what exists for aviation, nuclear power, food safety, and buildings. Such a framework would require Congressional action to create uniform national standards and to coordinate with other countries. The risk, he emphasized, was not merely domestic. Harms from AI systems developed in China, Asia, or Africa would reach American citizens just as readily as harms from American systems. Without coordinated oversight, the problem would only multiply.

Meta CEO Mark Zuckerberg, speaking after the White House signing, called the pledge a beginning. It was not, he said, the only thing the industry would ever do. It was a start. But Russell's skepticism cut deeper: voluntary commitments had failed before, in an industry that had repeatedly promised to do better while doing less. The FTC investigation suggested that at least some parts of the government were not content to wait and see whether this time would be different.