Industry insiders told Axios that the competition for AI dominance among labs, companies and nations may be fundamentally at odds with efforts to prioritise safety. While leading AI firms are publicly proposing more independent oversight and mechanisms to slow progress, the financial incentives and resources required to scale models create persistent pressure to keep accelerating development.
Why this matters
Top AI companies have been advocating for third-party evaluators and self-policing arrangements to justify a pause or slowdown. Critics argue those proposals clash with the enormous monetary rewards associated with continued rapid advancement, and with investor expectations tied to sky-high valuations.
Current state of play
As AI CEOs promote self-regulation and independent evaluation, deployed models are increasingly opaque about how they produce answers and solutions. The growing complexity of AI systems, together with the deployment of thousands of partly autonomous agents for different tasks, makes real-time auditing and review extremely difficult.
There is also a trust gap around independent safety groups because of close ties between industry actors and some evaluators, and because many researchers share a common pool of investors and funders within the AI research community.
Follow the money
The incentive to move fast is far more financially attractive than calls to slow down. Axios reports that OpenAI and Anthropic are preparing to go public with multi-trillion-dollar valuations, creating pressure to deliver progressively better models to justify those valuations.
Goldman Sachs projects more than $7 trillion will be spent on scaling AI over the next five years. Frontier labs will need to generate returns on that investment, which pushes them to outpace competitors. That race also includes efforts to prevent customers from switching to open-weight model providers, which can be cheaper to run and are regarded by some as a safer option.
Views from the field
Daniel Kokotajlo, a former OpenAI researcher and executive director of the AI Futures Project, said he does not believe AI can be safely scaled. Kokotajlo argued that top AI firms should avoid directing their vast computing power solely toward model improvement, which would require a slowdown in order to be safe. He added that researchers and employees involved in scaling "are doing something wrong."
Kokotajlo left OpenAI in 2024 over safety concerns and has since been invited to speak to current employees about his AI safety research. "The companies are just having this continuous process of conflicted feelings and internal discussion about like, what are we doing," he said. "Should we stop? Are we the good guys?"
Others remain more optimistic. Former President Donald Trump has said he will not stifle AI and framed dominance in AI as a win. Akshay Krishnaswamy, chief architect at Palantir, told Axios that "human willpower is capable of figuring out how to safely scale these models," and suggested that open models are one route to safer scaling.
Bottom line
Despite growing rhetoric about an AI slowdown and proposals for independent oversight, it remains unclear whether such a slowdown is achievable in the near term. Trillions of dollars of investment, market pressures, technical complexity, and questions around independent auditing create significant obstacles to placing safety decisively ahead of speed.
