Dario Amodei says AI companies should slow down. Critics say that suits them just fine.

Key points

• Dario Amodei, founder of Anthropic (the company behind the Claude chatbot), published an essay calling on tech companies to slow development of the most advanced AI models.
• Anthropic competes directly with OpenAI and Google in the race to build increasingly powerful AI systems.
• Senator Bernie Sanders and strategist Steve Bannon have both called for government restrictions on AI, though for very different reasons.
• The Guardian first flagged the tension between genuine safety concerns and the commercial advantages that scary-AI narratives can create for established players.

Dario Amodei wants the AI industry to hit the brakes. In an essay published Saturday, the Anthropic founder argued that tech companies are moving too fast on their newest, most powerful models and need to slow down. On its face, that sounds like a responsible call from someone who knows the technology well.

Here is the awkward part.

Anthropica already sits near the top of the AI industry. A company asking the whole field to slow down also happens to be asking its smaller, younger rivals to stop catching up. That conflict of interest is worth naming out loud. Our earlier story on 14 September found that critics are already calling the multi-CEO commitment to slow development a cartel move.

This is a pattern security and policy researchers call "safety-washing": wrapping competitive strategy in the language of existential risk. The risk may be real. Readers should hold both ideas at once.

What is Amodei actually worried about?

His concern is the newest generation of AI models, systems far more capable than the chatbots most people use today. He argues development is moving faster than anyone's ability to check whether these systems behave safely. That concern is shared by independent researchers who have no financial stake in slowing anyone down.

The problem is that "we might destroy humanity" is also an extraordinary marketing claim. Any company that frames itself as the one taking danger seriously benefits from the public believing the danger is real. Worth remembering: Amodei has already opened Anthropic's models to independent evaluators, as we reported on 12 September, which at least gives the safety argument some operational weight.

Who else is calling for restrictions?

The calls are coming from unexpected places. Senator Bernie Sanders and Steve Bannon have both called for government limits on AI. They agree on almost nothing else, but both described the situation as a "cold war" with China over who controls the technology. What that government action should actually look like is a much harder question, and neither figure offered a detailed answer.

What should ordinary readers make of all this?

Be a careful listener. When a company warns loudly that its own product could be dangerous, ask who benefits from that warning.

That doesn't mean dismiss the concern. Researchers outside the industry, with no revenue at stake, have raised similar alarms for years. The technology is genuinely powerful and the questions about its limits are genuinely unsettled.

But public fear is also a resource that well-funded companies can shape. The most useful habit right now is separating what independently funded researchers are saying from what companies with products to protect are saying.

Watch for: safety arguments that appear precisely when a competitor is gaining ground, calls to slow down that arrive without any concrete proposal for how, and government officials reaching for Cold War analogies before they have a clear policy to match.