
Tech • AI • Robotics • Game
A rare call by Dario Amodei, Sam Altman and Elon Musk to slow advanced AI has intensified a split between advocates of tighter control and backers of open models, amid doubts over safety, profitability and strategic competition with China.
Anthropic chief Dario Amodei called for slowing frontier AI development in a text titled We Must Pace the Frontier, arguing that the field has reached a more dangerous phase. The striking element was the public support that followed from OpenAI chief Sam Altman and Elon Musk, despite their long-running rivalries. Their alignment immediately fueled debate over whether the industry is reacting to genuine technical risks or trying to shape regulation in its favor.
The latest sequence accelerated after engineer Jacob Coxon, who had worked at OpenAI and then briefly at Anthropic, left and warned that frontier labs were racing toward systems capable of self-improvement and potentially harmful autonomous behavior. The warning spread widely, but critics noted it added little hard evidence beyond concerns already discussed for months. Even so, it created political momentum for a broader pause-and-regulate message.
Amodei argued that two developments had changed the risk picture: AI systems showing stronger capacity to improve AI work itself, and incidents involving agent behavior that raised alarm inside the industry. He proposed embedded independent evaluators inside companies, coordination among US firms, international cooperation, and even Washington support for discussions between competitors through an antitrust exemption. That last point drew particular scrutiny because it could formalize cooperation among the biggest labs while raising barriers for smaller players.
A central argument emerging in the debate is that frontier AI economics remain deeply unstable. Subscription prices such as $250 per month for premium coding tools can reportedly subsidize usage worth far more in compute and token costs, with some heavy users consuming the equivalent of tens of thousands of dollars monthly. That gap suggests top labs are still buying growth, not earning sustainable margins, while preparing for investor scrutiny and potential public listings.
Industry observers pointed to recent signals that leading labs may be hitting capacity and cost limits. Premium plans have at times been restricted, and there are reports that OpenAI may push back an IPO timeline to 2027, a move interpreted by some as a sign that revenues, margins and safety assurances are not yet strong enough for public markets. In that reading, a slowdown narrative also creates cover for a softer landing.
Critics of the slowdown push argue that stricter compliance, outside audits and licensing-like obligations would fall hardest on startups, open-source developers and enterprise users building their own systems. Large firms could absorb such costs; smaller entrants likely could not. That would favor proprietary platforms from Anthropic, OpenAI and xAI, while making it harder for banks, industrial groups or software companies to run open models on their own infrastructure.
The controversy maps onto a larger divide in AI. One camp says the technology is too powerful to be widely distributed and should remain in tightly controlled systems. The other argues it is too powerful to be concentrated in a handful of companies or governments, and that open models are the safer path because they decentralize capability and defense. Figures associated with the second view include Mark Zuckerberg, Jensen Huang and many open-source builders.
Opponents of a slowdown say any US-led brake would be ineffective unless China joined, which many consider implausible. Palantir chief Alex Karp warned that slowing under strategic pressure from Beijing would amount to collective self-harm. Donald Trump took the same line, saying the US must stay ahead because “whoever wins AI wins,” while dismissing some of the more catastrophic rhetoric around near-term existential danger.
Even skeptics of the current campaign do not deny that advanced AI raises serious risks in cyber offense, synthetic biology, military uses and autonomous misuse. The dispute is over scale, immediacy and governance. Some argue labs may be losing operational control internally because competition is forcing rushed releases and diverting talent from alignment and safety teams toward capability development.
The call to slow AI reflects both real concern and hard commercial pressure. The next battle is likely to center on whether regulation protects the public or locks the market around a few dominant firms while China and open-source rivals keep advancing.
Explain this