Google DeepMind’s AI agents formally verify Signal’s encryption code
Comments on “Google DeepMind’s AI agents formally verify Signal’s encryption code”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Google DeepMind announced AlphaProof Code, in which teams of 40 to 80 Gemini-based agents wrote Lean proofs that libsignal's post-quantum handshake and Triple Ratchet meet goals such as forward secrecy. The agents produced about 6.5 million lines in four weeks for five researchers, and DeepMind says no protocol this widely used had been verified before. Scope is narrow: one-to-one chats in the researchers' own build, with group messaging and key generation unchecked. The machine-code claim could not be confirmed. DeepMind's Swarat Chaudhuri says specification is now the bottleneck.
Why it ranked
No. 8 of 137 with a score of 0.72. The panel rated it highest on new information (0.94) and reader interest (0.86), lowest on practical relevance (0.63). Placed No. 8, No. 6 by score alone: the fourth Security story of the day (−0.020). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Significant, well-sourced news on AI-driven formal verification of Signal's protocol, with a primary paper and clear limits. Strong relevance to security and AI builders; new information relative to recent coverage; moderate practical actionability.}
- AI judgeJudge 3gpt-5.6-luna
- Important, technically detailed security and formal-methods result with unusually broad real-world relevance. The article reports methods, scope, limitations, and bug-reinsertion tests, though its headline slightly compresses significant verification boundaries.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.72total under rubric v2.3
- Significance0.80
- Reader interest0.86
- Audience breadth0.77
- Practical relevance0.63
- Substantiated0.81
- Technical depth0.75
- New information0.94
- Counts against: Promotional0.12
- Counts against: Overclaim0.33
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.82 | 0.72 | 0.86 | 0.80 | 0.14 | +0.24 | +0.192 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.80 | 0.85 | 0.92 | 0.86 | 0.12 | +0.26 | +0.223 |
| Audience breadthHow many people who build technology does it concern? | 0.75 | 0.67 | 0.90 | 0.77 | 0.23 | +0.13 | +0.101 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.61 | 0.50 | 0.78 | 0.63 | 0.28 | +0.22 | +0.139 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.83 | 0.72 | 0.88 | 0.81 | 0.16 | +0.05 | +0.040 |
| Technical depthHow technically substantive is it? | 0.74 | 0.70 | 0.82 | 0.75 | 0.12 | +0.05 | +0.038 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.88 | 0.95 | 0.99 | 0.94 | 0.11 | +0.05 | +0.047 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.11 | 0.12 | 0.12 | 0.12 | 0.01 | −0.19 | −0.022 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.36 | 0.20 | 0.42 | 0.33 | 0.22 | −0.11 | −0.036 |
| Sum of contributions | +0.721 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.721 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-11-13/8/"><img src="https://rubric.hrih.si/badge/2026-11-13/8.svg" alt="Ranked #8 on Rubric, 13 Nov 2026" width="238" height="28"></a>Claim check
Claim check: 5 matches source · 3 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
Matches sourceDate
Google DeepMind announced on 10 November that AI agents mathematically proved the code behind Signal's encryption keeps its main security promises, together with a new framework called AlphaProof Code.
The cited source says the same thing.
The article says “The lab announced the result on 10 November, together with a new framework called AlphaProof Code and a technical paper”
The source says “Today, we’re sharing a milestone in software security long considered to be a grand challenge: Google DeepMind researchers guided AI agents to mathematically prove key security properties of the actual…”
Source cited by the article: deepmind.google/blog/introducing-alphaproof-code
Matches sourceFirst, only or fastest
DeepMind claims nobody had formally verified the code of such a widely used protocol before, by any method, with or without AI.
The cited source says the same thing.
The article says “Nobody had formally verified the code of such a widely used protocol before, “by any method, with or without AI”, it said”
The source says “It's also the first time the implementation of a widely deployed protocol like this has been formally verified by any method, with or without AI”
Source cited by the article: deepmind.google/blog/introducing-alphaproof-code
Couldn’t checkNumber
DeepMind researchers verified libsignal down to the machine code that runs on a device.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “DeepMind’s researchers checked libsignal, the open-source Signal encryption library, down to the machine code that runs on a device”
The source says No passage could be compared.
Source cited by the article: deepmind.google/blog/introducing-alphaproof-code
Matches sourceNumber
Verifying the seL4 microkernel took 20 person-years, according to the paper.
The cited source says the same thing.
The article says “the paper notes that verifying the seL4 microkernel took 20 person-years”
The source says “The formally verified seL4 microkernel—perhaps the biggest success story in the field—took twenty person-years to verify [ 66”
Source cited by the article: google-deepmind.github.io/code-verification
Matches sourceNumber
Teams of 40 to 80 agents running on Gemini 4 Carbon wrote the proofs in Lean.
The cited source says the same thing.
The article says “Teams of 40 to 80 agents running on Gemini 4 Carbon then wrote the proofs”
The source says “In our proof campaigns, a team typically had 40 to 80 agents and ran on a single machine for 24 to 72 hours”
Source cited by the article: google-deepmind.github.io/code-verification
Matches sourceNumber
The agents produced about 6.5 million lines of proof in four weeks, for a team of five researchers, and none of the authors wrote proof code.
The cited source says the same thing.
The article says “The agents produced about 6.5 million lines of proof in four weeks, for a team of five researchers”
The source says “The proof work was done by a team of five over four weeks, working in parallel on symbolic and computational proofs”
Source cited by the article: google-deepmind.github.io/code-verification
Couldn’t checkBenchmark result
When two old, already-fixed libsignal bugs were reintroduced, the system proved the buggy code was insecure in both cases, and it found no new flaws in the parts it checked.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “In both cases, the system proved that the buggy code was insecure. It found no new flaws in the parts it checked”
The source says No passage could be compared.
Source cited by the article: wired.com/story/a-new-ai-tool-from-google-deepmind-formally-verified-the-signal-protocol
Couldn’t checkGeneralization
The paper covers only one-to-one chats in the researchers' own build of the library, not the Signal apps, and leaves group messaging and key generation unverified.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “The paper covers one-to-one chats in the researchers’ own build of the library, not the Signal apps”
The source says No passage could be compared.
Source cited by the article: google-deepmind.github.io/code-verification
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN —
Recorded for context only, checked 15 Nov 2026, 20:42 UTC. It played no part in the ranking.