What's Scarier Than Agents Taking over Internet? CEO Cartel Trying Take over AI

Dario Amodei of Anthropic urges slowing frontier AI development via embedded evaluators inside firms, democratic coordination of safety standards (which would require loosening antitrust rules), and attempts at global non‑proliferation. The piece argues those measures would effectively hand market power to the biggest players - Anthropic, OpenAI, xAI and allies - by letting them inspect rivals’ work, shield themselves from liability, and bar startups from competing. Elon Musk’s quick assent to model inspections is noted as evidence the proposal risks becoming a cartel. Global pacing is labeled unrealistic and unverifiable: bad actors and rival states would have strong incentives to cheat, so a voluntary slowdown would leave democracies exposed while authoritarian or nimble competitors push ahead.

The author contends the real motive is economic: fading subsidies, rising data‑center costs, and competition from open‑source and low‑cost Chinese models make a truce attractive to incumbents seeking relief and regulatory barriers. Rather than throttling progress, the recommended response is to build the best defensive and beneficial AI, create a White Hat Institute to vet prerelease models and harden systems, involve government where appropriate, and outlaw ransom payments that fuel cybercrime. Slowing AI via cartelization is portrayed as a self‑serving strategy that would weaken national security and delay lifesaving benefits.