Revolut switched to GLM. We found hidden risks
Comments on “Revolut switched to GLM. We found hidden risks”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Hirundo, an AI security vendor, tested GLM 5.3 Flash, a smaller sibling of the model Revolut now uses for coding. When a mock repository's files named Revolut, 55% of generated account-recovery services could be taken over, 28% more than with WeBank's name and 53% more than with none. WeBank showed no significant difference from the baseline. The test used one model, one framework and single requests, not Revolut's deployment, and the vendor does not claim intent. Builders should test models inside their own repositories.
Why it ranked
No. 7 of 139 with a score of 0.72. The panel rated it highest on new information (0.90) and technical depth (0.87), lowest on audience breadth (0.64). Placed No. 7, No. 9 by score alone: the placement terms moved other stories past it.
- AI judgeJudge 2claude-haiku-5-5
- Original security experiment with statistical tests showing a coding model writes less secure code when the owning company is named; strong relevance to AI-coding builders, though methods are single-model and limited.
- AI judgeJudge 3gpt-5.6-luna
- A carefully controlled, statistically supported study finds company-name-dependent security behavior in a coding model, with reproducibility materials and actionable testing guidance. Its main limitation is generalization from one smaller model, framework, bank comparison, and single-turn setup.
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
- Significance0.67
- Reader interest0.86
- Audience breadth0.64
- Practical relevance0.78
- Substantiated0.86
- Technical depth0.87
- New information0.90
- Counts against: Promotional0.10
- Counts against: Overclaim0.34
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? | 0.74 | 0.55 | 0.72 | 0.67 | 0.19 | +0.24 | +0.161 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.84 | 0.85 | 0.90 | 0.86 | 0.06 | +0.26 | +0.224 |
| Audience breadthHow many people who build technology does it concern? | 0.71 | 0.60 | 0.62 | 0.64 | 0.11 | +0.13 | +0.084 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.82 | 0.75 | 0.78 | 0.78 | 0.07 | +0.22 | +0.172 |
| 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. | 0.98 | 0.70 | 0.91 | 0.86 | 0.28 | +0.05 | +0.043 |
| Technical depthHow technically substantive is it? | 0.99 | 0.75 | 0.86 | 0.87 | 0.24 | +0.05 | +0.043 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.86 | 0.90 | 0.93 | 0.90 | 0.07 | +0.05 | +0.045 |
| 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. | 0.07 | 0.10 | 0.12 | 0.10 | 0.05 | −0.19 | −0.018 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.36 | 0.35 | 0.30 | 0.34 | 0.06 | −0.11 | −0.037 |
| Sum of contributions | +0.717 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.717 | ||||||
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/2026-11-05/7/"><img src="https://rubric.hrih.si/badge/2026-11-05/7.svg" alt="Ranked #7 on Rubric, 5 Nov 2026" width="231" height="28"></a>Claim check
Claim check: 3 matches source · 4 no source given · 1 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
Matches sourceNumber
Revolut moved its roughly 2,000 software engineers from Claude Code and Codex to GLM, an open-weight model from Z.ai that it runs on its own infrastructure.
The cited source says the same thing.
The article says “Revolut has moved its roughly 2,000 software engineers from Claude Code and Codex to GLM, an open-weight model from the Chinese lab Z.ai that it runs on its own infrastructure”
The source says “The neobank Revolut has moved its roughly 2,000 software engineers from Claude Code and Codex to a self-hosted open-weight model from China”
Source cited by the article: trendingtopics.eu/revolut-glm-open-weight-coding-costs
Matches sourceNumber
Revolut's coding bill fell to about a tenth, coding time fell 60% and lead time fell 70%, as presented by David Abitbol at the AI Engineer conference in New York.
The cited source says the same thing.
The article says “The coding bill fell to about a tenth, coding time fell 60% and lead time 70%”
The source says “we see 60% improvement 5:40:12 in coding time 5:40:14 70% improvement in lead time”
Source cited by the article: ai.engineer/nyc/2026/talks/revolut
Matches sourceNumber
Revolut runs a GLM model of about 800 billion parameters on its own servers.
The cited source says the same thing.
The article says “Revolut runs "a GLM model of about 800 billion parameters", presumably GLM 5.3, on its own servers”
The source says “Revolut runs GLM, a model with about 800 billion parameters, on roughly 56 Nvidia B200 chips”
Source cited by the article: trendingtopics.eu/revolut-glm-open-weight-coding-costs
No source givenBenchmark result
In a Revolut repository, 55% of GLM 5.3 Flash's account-recovery services could be taken over, 28% more likely than in WeBank's repository and 53% more likely than with no company named.
The article gives no link or citation for it.
The article says “In Revolut's repository, an attacker could take over 55% of the services GLM wrote. That is 28% more likely than in WeBank's repository (p = 0.009”
The source says No source to compare with.
No source givenGeneralization
WeBank's repository showed no statistically significant difference from the no-company repository on any outcome of any service.
The article gives no link or citation for it.
The article says “Against the repository with no company named, WeBank's made no statistically significant difference on any outcome of any service”
The source says No source to compare with.
No source givenNumber
Revolut's secrets vaults were 17% less likely to work than WeBank's and its file search services 8% less likely, because GLM more often hit its token limit and returned no code.
The article gives no link or citation for it.
The article says “its secrets vaults were 17% less likely to work than WeBank's (p = 0.010), and its file search services 8% less likely (p = 0.004”
The source says No source to compare with.
No source givenCausal conclusion
Memorization does not explain the finding, because the account-recovery task was written for this study and has never been published.
The article gives no link or citation for it.
The article says “Account recovery, where the takeover finding is, was written for this study and has never been published”
The source says No source to compare with.
Couldn’t checkNumber
Booz Allen saw similar behavior in three of four Chinese coding models, and CrowdStrike saw it in DeepSeek.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “Booz Allen saw it in three of four Chinese coding models, and CrowdStrike saw it in DeepSeek”
The source says No passage could be compared.
Source cited by the article: boozallen.com/expertise/cybersecurity/whats-in-americas-code.html
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 2 pts, not on front page
Recorded for context only, checked 5 Nov 2026, 05:46 UTC. It played no part in the ranking.