Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
81 76 87 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
Insufficient record: 2 predictions resolved, 5 needed for a score.
-
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
65 45 83
All types combined: 80 (likely 75 to 85, high confidence), from 84 claims: 78 factual claims, 6 promises. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Demis Hassabis's claims
42 claims
“Our AlphaProteo model can design novel, high-strength protein binders that target diverse molecules – including those associated with cancer and diabetes.”
Hassabis et al., Google DeepMind blog 25 November 2025 AlphaFold: Five Years of Impact (Google DeepMind)
AlphaProteo can design novel, high-strength protein binders targeting diverse molecules, including ones associated with cancer and diabetes.
Mostly true True in substance, wrong in a detail.
Said for an organisation they lead or work for; it counts on both records.
What would count as true or false
True: published AlphaProteo results (Google DeepMind's September 2024 blog post and technical report, or later peer-reviewed work) show the model designing novel protein binders with high binding affinity against several different target proteins, including at least one target linked to cancer (e.g. VEGF-A, PD-L1) and at least one linked to diabetes or its complications (e.g. TrkA, IL-17A, or another target the documentation associates with diabetes), as of November 2025. Mostly true: the results show novel, high-affinity binders against diverse targets including cancer-associated ones, but the link to diabetes is indirect or loosely stated (e.g. the target is linked to a complication of diabetes rather than to diabetes itself). Mixed: novel binders were designed, but the high-strength claim or the claimed range of targets is only partly supported. Mostly false: the binders were only computationally predicted without experimental validation of their strength, or they target only a narrow set of molecules. False: AlphaProteo does not exist or has not been shown to design binders as described.
Evidence
- AlphaProteo generates novel proteins for biology and health research Google DeepMind · 5 September 2024 Own record
“AlphaProteo can generate new protein binders for diverse target proteins, including VEGF-A, which is associated with cancer and complications from diabetes.”
- De novo design of high-affinity protein binders with AlphaProteo arXiv (authors from Google DeepMind) Own record
“We identified 9 total binders with sub-nanomolar \(\mathrm{K_{D}}\) values: 4 for IL-7RA, 2 for PD-L1, 1 for TrkA, and 2 for VEGF-A (Table S7).”
- De novo design of high-affinity protein binders with AlphaProteo alphaXiv · 12 September 2024
“Only eight targets were tested experimentally, all with a target crystal structure as input, and SC2RBD, PD-L1, and TrkA were used during method development, so their rates may overstate performance on new targets.”
- AlphaProteo generates novel proteins for biology and health research Google DeepMind · 5 September 2024 Own record
“However, we know that our AI system has limitations, as it was unable to design successful binders against an 8th target, TNFɑ, a protein associated with autoimmune diseases like rheumatoid arthritis.”
Reasoning
E2 reports measured sub-nanomolar KD binders for several targets including VEGF-A and PD-L1 (both cancer-associated), supporting novel high-strength binders against diverse targets. E1 names VEGF-A as linked to cancer and to complications of diabetes, which is an indirect diabetes link rather than a direct one. Under the criteria this indirect diabetes connection fits mostly_true. E3 and E4 note limits (few targets tested, one failure) but do not overturn the core result.
Cite this claim
Demis Hassabis, “AlphaProteo can design novel, high-strength protein binders targeting diverse molecules, including ones associated with cancer and diabetes.” (said 25 November 2025). Verdict: mostly true. Reliability API, claim c_ff4e09904668, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_ff4e09904668
Claims without a verdict
Is a verdict here wrong, or do you speak for Demis Hassabis? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.