The AI safety movement can't agree on what it's even trying to prevent, which makes it the perfect vehicle for everyone's favorite policy outcome: nothing.
The Summary
- AI safety has fractured into camps that disagree on fundamental risks: some worry about existential threats from superintelligence, others focus on present-day harms like bias and misuse
- Self-regulation by AI companies has become the default policy stance, which is another way of saying no regulation at all
- The tent is so big that "AI safety" now means everything and therefore nothing, paralyzing any coherent policy response
The Signal
The AI safety debate has a structural problem that no one wants to name. AI Snake Oil identifies an unstated disagreement fracturing the movement: one camp fears superintelligent AI wiping out humanity, while another worries about today's models amplifying bias, spreading misinformation, and concentrating power. These aren't different degrees of the same concern. They're different worldviews with incompatible policy prescriptions.
The existential risk crowd wants compute governance, model capability limits, and international coordination to prevent a hypothetical AGI catastrophe. The present-harm crowd wants transparency requirements, fairness audits, and accountability for damage happening right now. One group treats AI like an unexploded bomb. The other treats it like a poorly regulated pharmaceutical.
"The tent is so big that AI safety now means everything and therefore nothing."
Wired cuts through the noise: asking AI companies to self-regulate is theater. It's what you do when you want credit for taking action without the inconvenience of actually constraining anyone. The AI industry has learned from Big Tech's playbook. Promise voluntary commitments, create internal ethics boards, publish safety research that doubles as marketing, and wait for the regulatory moment to pass.
The timing matters. We're in the window where AI policy gets written, and the fracture in the safety movement is a feature for the companies, not a bug. When advocates can't agree on what they're protecting against, legislators default to the path of least resistance. That path runs straight through industry-drafted frameworks that sound serious and do nothing.
Key dynamics at play:
- Philosophical disagreements get weaponized as proof that "experts can't even agree"
- Companies fund both sides of the safety debate, ensuring neither wins
- The big tent becomes an excuse for small action
The result is a safety movement that can't set priorities because it can't agree on threats. Meanwhile, the policy vacuum gets filled with voluntary commitments that commit to nothing. The companies building the most powerful AI systems get to grade their own homework.
The Implication
If you care about AI governance, watch for two things. First, who's funding the research and advocacy on both sides of the safety debate. Follow the money and you'll understand why consensus stays out of reach. Second, track the gap between voluntary commitments and actual behavior. Self-regulation works great until it doesn't, and by then it's too late to write the rules.
The next twelve months will show whether the safety movement can unify around actionable policy or whether it stays big enough to accomplish nothing. The companies are betting on the latter. Prove them wrong or stop calling it safety.