ClaimsDario Amodei

“We believe AI for biology is on a similar exponential trend.”

Dario Amodei, Anthropic CEO, on X 23 September 2026 Post by @DarioAmodei on X, 2026-09-23 (X)

Prediction AI capabilities

Amodei said that "we" believe AI for biology is on an exponential trend similar to the one he described for AI in mathematics.

Mostly false False in substance, right in a detail.

Prediction: Statements about events outside the person's control, checked once they fell due: did they come true?

Said for an organisation they lead or work for; it counts on both records.

What would count as true or false

Judge the substance of the stated belief as of 2026-09-23: is AI capability on biology tasks improving on an exponential (compounding, roughly constant-doubling) trend comparable to the one Amodei described for AI in mathematics? Speaker hedging ('we believe') is noted, but the test is the trend itself. TRUE: benchmark time series, capability-tracking studies or expert assessments from around the statement date show AI performance on biology tasks (for example protein/structure prediction, biology QA or lab-protocol benchmarks, autonomous research tasks) rising at a compounding rate broadly comparable to the cited math trend. MOSTLY TRUE: the evidence shows clear compounding improvement in biology, but at a noticeably slower or less clean rate than math, or across only some major biology task families. MIXED: the evidence is split, with exponential-looking progress on some biology measures and flat or linear progress on others in roughly equal weight. MOSTLY FALSE: biology AI is improving, but the evidence shows no exponential trend comparable to math, or the claim rests on a narrow benchmark presented as the whole field, or no data establishes such a trend. FALSE: the evidence shows biology AI capability stalling or improving only marginally over the relevant period.

Set before the research began.

Evidence (4)

  1. E1 Overall trends SecureBio

    “Capabilities are rising over time, in both general biology and the biosecurity-relevant areas tested.”

  2. E2 Overall trends SecureBio

    “Companies that have added chemical, biological, radiological, and nuclear (CBRN) safeguards have published models that deliberately decline to answer many biology questions, pulling their scores below the upward trend.”

  3. E3 Frontier AI Trends Report by The AI Security Institute (AISI) UK AI Security Institute

    “At the beginning of 2024, for the first time, models performed better than experts (biology PhDs) on our open-ended biology questions. Since then, we have observed continuous improvements on these question sets.”

  4. E4 Introducing GeneBench-Pro OpenAI · 30 June 2026

    “That is a sharp increase from when we began building the original GeneBench; at that time, our best frontier model, GPT‑5, scored below 5%.”

Reasoning

The evidence shows biology AI capability rising. SecureBio reports capabilities rising over time in general biology and biosecurity-relevant areas (E1), the UK AI Security Institute reports continuous improvement since models passed biology PhD experts in early 2024 (E3), and OpenAI reports a sharp rise on GeneBench-Pro (E4). However, none of these sources gives a doubling time or an exponential fit for biology. E4 compares a single benchmark result against one earlier figure, and E2 notes that refusals under CBRN safeguards pull some models' scores below the upward trend. The criteria say mostly false when biology AI is improving but no data establishes an exponential trend comparable to the one cited for math, and that is what this evidence shows.

Cite this claim

Dario Amodei, “Amodei said that "we" believe AI for biology is on an exponential trend similar to the one he described for AI in mathematics.” (said 23 September 2026). Verdict: mostly false. Reliability API, claim c_6e5260655f16, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_6e5260655f16

This claim is in dataset release v1.1.2, with the same verdict.

Reply to this claim

For Dario Amodei, or someone speaking for them: a correction or counter-evidence. Replies are read before anything is shown; an approved reply shows your name, role, date, statement and link. Your email address is never shown and is deleted 30 days after the decision (privacy).

More prediction claims by Dario Amodei

  1. Resolved true Prediction 28 February 2025
    “for coding, we're going to see very serious things by the end of 2025,”

    By the end of 2025, AI will do very serious things in coding.

  2. Resolved false Prediction 11 November 2024
    “But I would guess that in another 10 months, we’ll probably get pretty close. We’ll be at least 90%.”

    Amodei guessed that within about 10 months AI models would probably get pretty close on typical real-world programming tasks, reaching at least 90%.

  3. Mostly false Prediction 11 November 2024
    “in 10 months we’ve gone from 3% to 50% on this task. And I think in another year we’ll probably be at 90%.”

    Amodei said that, after progress from 3% to 50% on SWE-bench in 10 months, state-of-the-art models would probably be at about 90% in another year, while adding that he did not know and it might be less.

  4. Mostly true Prediction 11 November 2024
    “I would guess a year from now, the models can do this very, very reliably.”

    Amodei guessed that by about November 2025, AI models would be able to perform computer-use tasks very, very reliably.

  5. Mostly true Prediction 28 February 2026
    “We're gonna be fine. The-- the impact of this designation is fairly small.”

    Anthropic will be fine, and the impact of the supply chain risk designation on it will be fairly small.

All of them

Verdicts are made automatically; how verdicts are made.