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?
72 65 79 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
66 54 78 -
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 1 promise resolved, 5 needed for a score.
All types combined: 71 (likely 64 to 77, high confidence), from 79 claims: 60 factual claims, 19 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Balaji Srinivasan's claims
37 claims
“For example, AI can break captchas, but it can’t fake onchain balances.”
Balaji Srinivasan, author of his own blog post 3 August 2025 AI is polytheistic, not monotheistic (balajis.com)
AI can break captchas but cannot fake onchain balances.
Mostly true True in substance, wrong in a detail.
The second judge said resolved true; the verdict shown is the first judge's.
What would count as true or false
True: as of 2025-08-03, AI systems can solve captchas, and AI cannot fake balances recorded onchain, meaning it cannot produce a false balance that passes checks against the real blockchain state without compromising keys or consensus. Mostly true: both parts hold in substance, with a small caveat (for example, AI can fool a naive interface or a forged screenshot, but not a verified onchain balance). Mixed: one part holds and the other is materially wrong. Mostly false: both parts are substantially overstated. False: both parts are shown to be wrong.
Evidence
- Breaking reCAPTCHAv2 arXiv (ar5iv render); accepted at COMPSAC 2024 · 20 February 2024
“Our main result is that we can solve 100% of the captchas, while previous work only solved 68-71%.”
- Are {CAPTCHAs} Still Bot-hard? Generalized Visual {CAPTCHA} Solving with Agentic Vision Language Model USENIX Association (USENIX Security 25) · 1 November 2024
“which achieves a solving rate of 60.7% on 2,600 challenges belonging to 26 types of visual CAPTCHAs.”
- State Root: Blockchain's Cryptographic State Commitment ChainScore Labs (glossary) · 3 November 2026
“acts as a tamper-evident commitment, allowing any network participant to cryptographically verify that a specific piece of state data (e.g., an account's ETH balance) is correct and current”
- onchain-postmortems/verus-ethereum-bridge-forged-proof/README.md at main · RealSpap/onchain-postmortems GitHub (RealSpap/onchain-postmortems repository) · 17 May 2026
“an attacker forged a cross-chain proof root and got the bridge's own Ethereum contract to release funds against an export that never existed on the Verus chain”
Reasoning
The first part holds. Independent research shows automated systems solving captchas: E1 reports a 100% solve rate on reCAPTCHAv2 image challenges, and E2 reports an agentic vision-language model solving 60.7% of 2,600 challenges across 26 visual captcha types. On the second part, E3 says the state root lets any network participant cryptographically verify an account's balance against the real chain, which supports the claim that a fake balance cannot pass checks against actual blockchain state. One caveat applies: E4 describes a forged proof that got a bridge contract to release funds, so a weak verifier can still be fooled. That attack did not fake the real chain state, and E4 does not say AI was involved, so I rate the claim mostly true rather than fully true.
Cite this claim
Balaji Srinivasan, “AI can break captchas but cannot fake onchain balances.” (said 3 August 2025). Verdict: mostly true. Reliability API, claim c_50717e581f0d, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_50717e581f0d
Claims without a verdict
- Unresolvable 148
- Not yet due 21
- Own records only 2
- Not researched 227
- Not checkable 1,091
Is a verdict here wrong, or do you speak for Balaji Srinivasan? 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.