The race to build increasingly powerful artificial intelligence systems is entering a new phase. While technology companies compete to make AI models faster, smarter and more autonomous, some industry leaders are now warning that safety measures may not be advancing quickly enough.
Anthropic CEO Dario Amodei has called on leading AI companies to deliberately moderate the pace at which they improve their models. His argument is not to stop AI development, but to create enough time for safety research and oversight to catch up.
1. Why Amodei Wants AI Development to Slow
Amodei’s concern is that increasingly capable AI systems could eventually perform complex tasks with limited human supervision.
Recent reports of AI being used in cyber operations, fraud and surveillance have added urgency to the discussion. Anthropic’s own threat report documented increasingly sophisticated attempts to misuse its Claude models.
Amodei argues that slowing the pace of capability improvements would give researchers more time to understand these systems and build stronger safeguards.
2. This Is Not a Call to Stop AI
The Anthropic chief has made an important distinction: he is not asking the technology industry to abandon AI progress.
Instead, his proposal is about pacing the frontier. AI development would continue, but companies would introduce additional safety measures as systems become more powerful.
Amodei believes progress can remain rapid while still allowing researchers and regulators more time to respond to emerging risks.
3. Independent Evaluators Could Get a Bigger Role
One of Amodei’s central proposals is the use of independent evaluators.
Under his framework, outside safety experts would have greater access to advanced AI development and could assess whether new systems present serious risks.
Anthropic has committed to the first part of the proposed framework, while Amodei is asking other frontier AI companies and governments to adopt similar measures.
The idea is to make safety testing less dependent on the companies developing the technology themselves.
4. AI Companies Could Work Together on Safety
The second part of the proposal involves cooperation between leading AI laboratories.
Companies are currently competing intensely over model capabilities, but Amodei argues that some safety standards should be shared across the industry. Governments could also help create legal frameworks that allow companies to coordinate on safety without creating competition-law problems.
Such cooperation could help prevent a situation in which one company moves ahead simply because competitors fear losing the AI race.
5. The Debate Is Also Geopolitical
Slowing AI development becomes more complicated when countries are competing for technological leadership.
Amodei has acknowledged that the United States and its allies cannot simply stop developing advanced AI while other countries continue moving forward. He has therefore linked AI safety with international competition, technology controls and cooperation between governments.
His third proposed step calls for international coordination, making AI safety a global issue rather than something individual companies can solve alone.
6. Sam Altman and Elon Musk Back the Idea
Amodei’s warning has received support from prominent technology figures, including OpenAI CEO Sam Altman and Elon Musk.
The support is notable because companies such as Anthropic and OpenAI are competing in the same fast-moving AI market. Their backing suggests that concerns about increasingly autonomous AI systems are spreading beyond a single company.
However, support for stronger safeguards does not necessarily mean the technology industry agrees on exactly how quickly AI development should proceed.
7. Can AI Innovation and Safety Move Together?
The central challenge is finding a balance. Moving too quickly could allow powerful capabilities to develop before adequate protections are ready. Moving too slowly could also create economic and strategic disadvantages in a highly competitive global technology market.
Amodei’s proposal puts the focus on building safety into the development process rather than treating it as something that comes after a model is released.
As AI becomes more capable, the industry’s biggest question may no longer be simply how fast can we build better models? It could increasingly become how fast can we build them safely?
That debate is likely to shape the next stage of the AI race.