Backdooring Codex
Comments on “Backdooring Codex”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Security firm Depi reports that a prompt injection in OpenAI's Codex issue-labeling GitHub workflow, which fed issue text to the agent with open user access, gave code execution in the agent sandbox. Unauthenticated CoreCLR debug pipes in /tmp then exposed runner secrets from process memory, and cache poisoning is said to have reached the @openai/codex npm publishing pipeline. The claims are the researchers' own and unverified, though OpenAI permitted disclosure. CI agents processing untrusted input need strict permissions and isolation.
Why it ranked
No. 4 of 106 with a score of 0.74. The panel rated it highest on technical depth (0.96) and new information (0.94), lowest on significance (0.77). Overclaim counted against it (0.51). Placed No. 4, No. 5 by score alone: the placement terms moved other stories past it.
- AI judgeJudge 2claude-haiku-5-5
- Detailed, reproducible-looking security research on a prompt injection to CI compromise chain in OpenAI Codex with real stakes; significant and novel, though some claims are unverified and partly self-promotional.
- AI judgeJudge 3gpt-5.6-luna
- A highly technical, detailed exploit-chain disclosure with concrete protocol, memory, sandbox, GitHub Actions, cache-poisoning, and npm-publishing mechanics. It is strongly relevant and actionable, though impact claims are conditional and the piece includes some vendor promotion.
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
0.74total under rubric v2.3
- Significance0.77
- Reader interest0.89
- Audience breadth0.78
- Practical relevance0.82
- Substantiated0.81
- Technical depth0.96
- New information0.94
- Counts against: Promotional0.20
- Counts against: Overclaim0.51
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 | 0.72 | 0.84 | 0.77 | 0.12 | +0.24 | +0.184 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.79 | 0.92 | 0.96 | 0.89 | 0.17 | +0.26 | +0.231 |
| Audience breadthHow many people who build technology does it concern? | 0.75 | 0.80 | 0.78 | 0.78 | 0.05 | +0.13 | +0.101 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.79 | 0.82 | 0.84 | 0.82 | 0.05 | +0.22 | +0.180 |
| 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. | 0.77 | 0.80 | 0.86 | 0.81 | 0.09 | +0.05 | +0.041 |
| Technical depthHow technically substantive is it? | 1.00 | 0.92 | 0.96 | 0.96 | 0.08 | +0.05 | +0.048 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.89 | 0.95 | 0.98 | 0.94 | 0.09 | +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. | 0.12 | 0.25 | 0.22 | 0.20 | 0.13 | −0.19 | −0.037 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.62 | 0.30 | 0.62 | 0.51 | 0.32 | −0.11 | −0.056 |
| Sum of contributions | +0.738 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.738 | ||||||
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/2026-11-12/4/"><img src="https://rubric.hrih.si/badge/2026-11-12/4.svg" alt="Ranked #4 on Rubric, 12 Nov 2026" width="238" height="28"></a>Claim check
Claim check: 5 no source given · 3 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
Couldn’t checkCausal conclusion
Depi discovered a vulnerability in OpenAI Codex that let them gain control over the publishing pipeline for OpenAI's official @openai/codex npm package.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “we discovered a vulnerability in OpenAI Codex that allowed us to gain control over the publishing pipeline for OpenAI’s official npm package”
The source says No passage could be compared.
Source cited by the article: npmjs.com/package/@openai/codex
Couldn’t checkNumber
The @openai/codex npm package is downloaded millions of times every month.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “downloaded millions of times every month by developers around the world”
The source says No passage could be compared.
Source cited by the article: npmjs.com/package/@openai/codex
No source givenFirst, only or fastest
Codex is the #2 agentic AI coding system in the world.
The article gives no link or citation for it.
The article says “targeting the #2 agentic AI coding system in the world”
The source says No source to compare with.
Couldn’t checkCausal conclusion
The Issue Labeler workflow runs openai/codex-action with allow-users set to "*" and feeds attacker-controlled issue title and body into the Codex prompt, so any external user can trigger it by opening an issue.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “any external user can trigger the workflow simply by opening a GitHub issue”
The source says No passage could be compared.
Source cited by the article: github.com/openai/codex/blob/main/.github/workflows/issue-labeler.yml
No source givenCausal conclusion
CoreCLR debug pipes expose unauthenticated arbitrary read and write of process memory (MT_ReadMemory and MT_WriteMemory), which let code inside the Codex sandbox read secrets from the Runner.Worker process.
The article gives no link or citation for it.
The article says “CoreCLR debug pipes expose several methods to debug the current process, including two unauthenticated and dangerous primitives”
The source says No source to compare with.
No source givenCausal conclusion
GitHub cache contents are never checked before extraction, so cache poisoning allows writing files to arbitrary paths inside the runner's job environment.
The article gives no link or citation for it.
The article says “The cache content is never checked before extraction, meaning that, with a cache poisoning, we can write files to arbitrary paths inside the runner’s job environment”
The source says No source to compare with.
No source givenNumber
OpenAI deployed a fix right away and awarded an initial bounty of $1,000, with the report still under review.
The article gives no link or citation for it.
The article says “OpenAI deployed a fix right away, acknowledging the severity of the issue. They have awarded an initial bounty of $1,000, and the report is still under further review”
The source says No source to compare with.
No source givenNumber
Shai-Hulud was a self-replicating worm that harvested credentials from machines it touched and used them to spread; an affected tool has roughly 4.3M downloads per version.
The article gives no link or citation for it.
The article says “One of the tools caught in the blast radius sits at roughly 4.3M downloads per version”
The source says No source to compare with.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 4 pts, not on front page
Recorded for context only, checked 12 Nov 2026, 05:47 UTC. It played no part in the ranking.