Unconfirmed: Anthropic Insider Expects AI to Match Humans at AI Research in 9 Months

Dwarkesh Patel says an Anthropic employee told him two days earlier that they expect AI to match humans at AI research within nine months. Per a Reddit post summarizing his remarks, the employee based that on inside information they can't share. This is unconfirmed: one unnamed person, one private conversation, no documents.

What is known

Patel also said labs should make that evidence public. His frustration, in his words: "I think there’s a very frustrating element to the current situation right now, where people inside the AI labs claim to have evidence that makes them more worried. And they are not allowed to talk about it."

Separately, Crypto Briefing published a piece on October 15 by Diego Almada Lopez under the headline "Anthropic predicts Claude will surpass human researchers in nine months." It cites numbers Anthropic has put out:

• On September 17, 2026, Anthropic disclosed that Claude models lead 26% of its internal AI R&D work, up from under 1% in February 2026.
• Anthropic says more than 90% of measured R&D tasks involved AI collaboration at AL3 or higher as of August 2026.
• It reported no fully autonomous work (AL5) and about 30,000 active agents on its internal platform.
• It projects Claude could lead around 80% of internal AI R&D at AL4 (the company's designation for AI leading the work) by the end of 2026, if the trend continues.

What is not known

The two accounts don't match. Crypto Briefing frames nine months as Anthropic's own prediction, backed by public metrics. Patel's account is a single employee's view, backed by undisclosed internal evidence, according to the Reddit account. The Crypto Briefing report does not name an Anthropic statement that endorses nine months.

We also don't know who the employee is, what the evidence is, or whether Anthropic's leadership shares the timeline. The reporting includes no denial and no response from Anthropic to Patel's account.

The public numbers are a different measure from "matching human researchers." Leading 26% of R&D work is not the same as equaling a human researcher. Crypto Briefing itself notes that early 2026 gains by Claude agents on alignment tasks were sometimes not replicable in production models.

An 80% projection for AL4 by year-end is a trend line. Nine months is a rumor. Treat them accordingly.