“Over and over we discover drugs with AI that work on trials in animals but not safely enough in humans; we just don’t understand biology well enough.”
Our reading, the sentence that was checked:
Over and over, drugs discovered with AI work in animal trials but are not safe enough in humans, because we just don't understand biology well enough.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, F Cancer, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-f-cancer-f93b9b-a2
We recorded 7 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2026-03-16. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
From the speaker
- Over and over
- discover drugs with AI
- work on trials in animals
- not safely enough in humans
- we just don’t understand biology well enough
Our interpretation
- The claim concerns repeated failures across distinct AI-discovered drug candidates, not merely adverse events in several patients receiving one drug.
- “Not safely enough” means human safety or tolerability prevented viable development or use at an effective dose. An efficacy failure or commercial cancellation does not suffice.
- The final clause attributes these failures to inadequate biological understanding.
- The statement describes cumulative experience through publication, without supplying a starting date.
- The wording concerns AI drug discovery generally, although the surrounding essay discusses cancer.
Counts as supported if: Multiple documented candidates show animal efficacy followed by human safety failure, with investigations supporting inadequate biological understanding as the cause.
Counts as contradicted if: A sufficiently complete account of relevant candidates and failure causes shows that the asserted recurring animal-efficacy-to-human-toxicity pattern did not occur, or that its claimed causal explanation is wrong. Individual successes alone do not disprove recurrence.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
BenevolentAI reported on April 5, 2023 that BEN-2293 met its safety endpoint but missed its overall secondary efficacy endpoints [e1]. Rentosertib's 12-week trial completed follow-up in August 2024; results published June 3, 2025 reported treatment discontinuations related to liver toxicity or diarrhea but concluded that further investigation was warranted [e2, e3]. These records do not establish repeated cases of animal efficacy followed by human safety failure caused by inadequate biological understanding [e1, e2, e3]. Their limited candidate coverage also cannot rule out such failures elsewhere [e1, e2, e3].
Limitations
- The packet lacks candidate-level records establishing the complete claimed sequence and its biological cause.
- The cited trials do not establish long-term safety or provide a comprehensive accounting of AI-discovered drug failures.
- Several potentially relevant primary records were inaccessible.
- The claim supplies no numerical threshold for recurrence or precise definition of AI involvement.
Evidence (3 items)
Context only: BenevolentAI,
safe and well tolerated
Why it matters here: The disclosed Phase IIa results met the safety endpoint while missing the overall secondary efficacy endpoints. This is counterevidence to treating AI-related trial disappointment as synonymous with human toxicity, but one candidate cannot disprove recurring failures elsewhere.
- Publisher
- BenevolentAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Nature Medicine,
safe
Why it matters here: The AI-discovered candidate had a favorable short-term trial conclusion, although the abstract also reports treatment discontinuations associated with liver toxicity or diarrhea. Neither a favorable conclusion nor individual adverse events establishes or refutes the claimed recurring development-failure pattern.
- Publisher
- Nature Medicine
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: Insilico Medicine,
August 2024.
Why it matters here: This dates the trial underlying e2 before Marcus's post. The release also identifies generative AI involvement in target identification and molecular design.
- Publisher
- Insilico Medicine
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
Primary records documenting multiple AI-discovered candidates with animal efficacy followed by human safety failures that prevented viable development or use, with investigations supporting the asserted biological cause; or a sufficiently complete candidate-level accounting showing that the recurring pattern did not occur or its causal explanation is wrong, through March 16, 2026.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, F Cancer
Over and over we discover drugs with AI that work on trials in animals but not safely enough in humans; we just don’t understand biology well enough.
- Where
- paragraph 18
- Published
- 2026-03-16
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-f-cancer-f93b9b-a2: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.