Ex-Google safety chief Tom Siegel warns AI endangers children

By Splitfeed EditorialLast updated Sep 26, 2026, 20:18 UTC

On September 22, 2026, Tom Siegel, a former safety chief at Google, issued a public warning that artificial intelligence poses greater risks to children than social media did. He described the potential harm as "very severe" and specifically cited suicide and mental health risks as among his primary concerns. Siegel called for a slowdown in AI development and proposed mandatory crash tests for AI products aimed at or accessible to children, borrowing the automotive safety metaphor to argue that products reaching young users should be stress-tested for harm before deployment. No response from Google or any AI company was reported.

The warning carries weight because of Siegel's direct experience overseeing safety at Google during the period when the harms of social media to young people became a defining policy and legal battleground. Platforms including Instagram and TikTok faced congressional hearings, lawsuits, and mounting evidence linking heavy use to depression and self-harm among teenagers. Siegel's claim is that AI represents a steeper escalation of those same dynamics, not merely a continuation of them. What is genuinely in dispute is whether that comparison holds, since AI products differ substantially from social media feeds in how they interact with users, and whether a development slowdown is a realistic or effective policy lever.

With no published reactions from the Pro-AI, Anti-AI, or Middle Ground camps as of this writing, what each would typically argue can be sketched from their broader patterns. The Pro-AI camp would likely contend that child safety concerns are better addressed through targeted product regulation and parental controls than through broad slowdowns, and that AI also carries significant benefits for education and mental health support. The Anti-AI camp would be expected to treat Siegel's credentials and specificity as validation of longstanding calls for precautionary limits on AI deployment, particularly where vulnerable populations are involved. The Middle Ground camp would typically push for the crash-test framework itself as a workable compromise, separating the question of speed from the question of safety standards.

The proposal most worth watching is whether Siegel's crash-test concept finds its way into regulatory language at the OECD level, where his warning was first published, or into pending child safety legislation in the United States or European Union. Any formal rulemaking process or legislative hearing that takes up mandatory pre-deployment safety assessments for AI products accessible to minors would be the clearest signal that his argument is gaining institutional traction.