Anthropic CEO Dario Amodei warns next year is highly uncertain

Anthropic CEO Dario Amodei says the events of next year are highly uncertain. Coming from the person running one of the world’s most closely watched AI labs, that is not a throwaway line.

The comment lands at a delicate moment. Fresh research suggests AI may already be speeding up AI research. Anthropic is also reportedly preparing for a major IPO while carrying heavy long-term computing obligations.

The data behind the uncertainty

Much of the unease traces back to a report from METR, an AI evaluation group. Its November 2026 analysis of Anthropic’s internal data found that AI is boosting research and development to roughly double its historical pace.

The estimate comes with wide error bars. METR puts the possible range of acceleration anywhere from 1.3x to 3x.

That spread matters because of where it could lead. According to the research, faster AI-assisted R&D could bring meaningful progress toward recursive self-improvement, or RSI, by 2027. RSI describes AI systems that help build better versions of themselves, which then help build even better versions.

The research also points to possible superhuman AI research capabilities arriving by 2027, earlier than prior forecasts had suggested.

Outside forecasters are moving in the same direction. The AI Futures Project updated its predictions in November 2026. Its median expectation now places the arrival of a superhuman AI researcher in 2027, absent deliberate efforts to slow things down.

Then there is Anthropic’s own paperwork. Its S-1 filings from early November 2026 suggested the company might soon meet the threshold in its Responsible Scaling Policy tied to accelerated AI progress. The Responsible Scaling Policy is Anthropic’s internal rulebook for when stronger safety measures must kick in.

A CEO who has been asking to tap the brakes

Amodei has been signaling caution for a while. In September 2026, he published an essay arguing for a slower pace of improvement in AI model capabilities.

His reasoning was that safety and alignment work needs room to catch up with raw capability gains. Alignment, in plain terms, is the effort to make sure AI systems actually do what humans intend, rather than something adjacent and unwelcome.

In the essay, he cautioned against moving too fast toward systems capable of large-scale cyber risks. He also pointed to greater existential threats if advanced AI arrives before the guardrails do.

The money side of not knowing

Amodei has argued that misjudging the timing of data center investments by even a few years can be “ruinous” for businesses. The problem is that revenue is unpredictable, while the bills for compute infrastructure are very predictable indeed.

Anthropic’s own numbers show why the stakes are high. Estimates of the company’s recent annualized revenue run rate range between $65 billion and $81 billion.

All of this sits alongside the company’s preparation for a significant IPO and its substantial long-term computing obligations.

What this means

For investors watching the AI trade, the most important signal is the mismatch Amodei is describing. Capability timelines may be accelerating, but nobody can say precisely how fast, and revenue may not move in lockstep with the hardware spending required to stay competitive.

For Anthropic specifically, the IPO process creates an unusual dynamic. The company is asking public investors to buy into its growth story while its CEO argues for slowing the capability gains that partly drive that story.

On the safety side, the key things to watch are concrete. Does Anthropic formally cross its Responsible Scaling Policy threshold, and what safeguards follow if it does? Do METR’s acceleration estimates narrow toward the low end or the high end of that 1.3x to 3x range? And do forecasters like the AI Futures Project keep pulling their timelines forward?