Anthropic credits Claude with three of the five steps in a gene-editing discovery

Two of the five steps in Amodei's account of the discovery belong to people. Anthropic's life sciences team suggested the research area, and the team carried out the experiments; Claude read the literature and genome data, found something, and proposed the experiments to check it ^[4][11]. The verification was done in-house, by the team that announced the result ^[4]. Amodei's hedge is his own: "Its precise function, biotechnological utility (if any), or level of significance is not yet clear, but at minimum it is work I would have been proud to do as a PhD student," he wrote on X ^[3].

The five-to-ten-year cure claim rests on a pattern in a different subject. He said models struggled with average high-school mathematics in 2023, did well on competitions for the best high-schoolers in 2024, solved minor open problems in 2025, more significant ones in early 2026, and in late 2026 are beginning to solve "the top few open problems in all of mathematics" ^[5]. Five stages, four calendar years ^[12]. "We believe AI for biology is on a similar exponential trend," he said ^[5].

The expensive half of drug development is untouched, and Amodei says so. Faster discovery identifies new drug targets and produces more promising candidates for testing, he said, but it does not shorten clinical trials, and any resulting treatment still needs standard testing and regulatory review ^[8]. More candidates at unchanged trial length means more spending years before any of it returns. Checking a discovery is quicker: he said humans working with the model can validate key results in a few weeks ^[14].

On the step that would compress the loop further, he was explicit: "Eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren't doing that today," he said ^[7]. Anthropic is hiring instead. He said the company aims to grow its life sciences team and to work with other scientists to extend the approach to a broad range of problems ^[10]. The Fox Business account of the announcement does not include a licensing partner or a revenue figure ^[15]. Earlier the same day, Amodei spoke before the U.N. Security Council about international standards for managing the threats posed by rapid AI development ^[13].

In my view one finding of unclear function does not support a valuation, and a buyer repricing AI drug discovery this week is paying for Amodei's schedule. The counter is one he raised himself: some argue that biology's need for experiment limits what AI can do there, and he said that suspicion is wrong ^[6]. If an outside group characterises the molecular machine and the same team produces a second and third find within a few quarters, the exponential comparison has evidence under it. If the function is still unclear a year from now, the record is one unexplained molecular machine ^[2].