Edition of Saturday 21 November 2026

Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.

Comments on “Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.”

Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked

HD Moore of Dragos sent nearly two million prompts through AI routers and inference resellers, each seeded with a unique fake secret such as AWS keys or private URLs, and tracked where they surfaced. Three providers appear to have been breached, with stolen keys used tens of thousands of times, and some resellers logged prompts. A Chinese lab's scraper fetched URLs from over half its prompts, including many marked Zero Data Retention. For builders, privacy flags and router defaults are not reliable; choose upstream providers deliberately.

Read the original at arstechnica.com. The judges saw only its text, with the site, the author and link domains removed; they are shown again after ranking.

Why it ranked

No. 4 of 118 with a score of 0.78. The panel rated it highest on reader interest (0.90) and practical relevance (0.85), lowest on new information (0.56). The judges disagreed most on new information (spread 0.82), so the story is marked contested. Placed No. 4, No. 2 by score alone: the fourth Security story of the day (−0.020), and close to “Three breaches, endless leaks: Your AI supply chain needs canaries”, No. 17 on Friday 20 November 2026 (−0.049). The placement terms move stories; they are not part of the score.

AI judgeJudge 2claude-haiku-5-5
Strong original canary experiment with concrete evidence of credential theft, provider breaches, and privacy-policy violations across AI routers; highly relevant to builders using AI APIs, though methodology details are summarized from a secondary report.
AI judgeJudge 3gpt-5.6-luna
A well-supported report of large-scale canary testing exposing prompt theft, compromised providers, reseller logging, and privacy-control failures. Highly actionable for AI security and architecture, but substantially overlaps a recently covered AI supply-chain canary story.

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

0.78total under rubric v2.3Contestedjudges differ by 0.82 on new information

Judges’ mean answer per question, on a scale from 0 to 1. Questions marked minus count against the score.
  • Significance0.77
  • Reader interest0.90
  • Audience breadth0.83
  • Practical relevance0.85
  • Substantiated0.77
  • Technical depth0.72
  • New information0.56
  • Counts against: Promotional0.04
  • Counts against: Overclaim0.24

Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).

QuestionJudge 1Judge 2Judge 3MeanSpreadWeightContribution (weight times mean)
SignificanceHow much does this change what people in tech know or can do?0.760.720.820.770.10+0.24+0.184
Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on?0.860.900.950.900.09+0.26+0.235
Audience breadthHow many people who build technology does it concern?0.900.800.780.830.12+0.13+0.107
Practical relevanceHow useful is it to someone who builds software or tech products?0.810.850.880.850.07+0.22+0.186
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.0.670.850.800.770.18+0.05+0.039
Technical depthHow technically substantive is it?0.760.720.680.720.08+0.05+0.036
New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)?0.700.900.080.560.82 , above the contested threshold+0.05+0.028
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.0.060.030.020.040.04−0.19−0.007
OverclaimDo the headline or conclusions claim more than the body supports?0.330.200.180.240.15−0.11−0.026
Sum of contributions+0.782
Claim-check penalty0.000
Total, as published0.782

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.

Ranked #4 on Rubric, 21 Nov 2026
<a href="https://rubric.hrih.si/editions/2026-11-21/4/"><img src="https://rubric.hrih.si/badge/2026-11-21/4.svg" alt="Ranked #4 on Rubric, 21 Nov 2026" width="238" height="28"></a>

Claim check

Claim check: 8 matches source. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.

  1. Matches sourceNumber

    HD Moore's experiment sent nearly two million AI prompts containing unique, trackable secrets to AI providers and tracked where the secrets surfaced.

    The cited source says the same thing.

    The article says

    “We sent nearly two million AI prompts containing unique, trackable secrets to AI providers, then watched where those secrets surfaced”

    The source says

    “We sent nearly two million AI prompts containing unique, trackable secrets to AI providers, then watched where those secrets surfaced”

    Source cited by the article: dragos.com/blog/ai-supply-chain-security-canary-tokens

  2. Matches sourceNumber

    The experiment found evidence that at least three AI providers were compromised by attackers who siphoned the prompts they received.

    The cited source says the same thing.

    The article says

    “Evidence that at least three AI providers were compromised by attackers who used their access to siphon prompts they received”

    The source says

    “We found attackers actively stealing data from three different hosting providers”

    Source cited by the article: dragos.com/blog/ai-supply-chain-security-canary-tokens

  3. Matches sourceNumber

    Attackers who compromised one provider reused a couple hundred of the seeded AWS keys, and every request sent during the late October windows was stolen.

    The cited source says the same thing.

    The article says

    “Each key was sent in exactly one request to one provider, all within a few short windows in late October, and every request we sent during those windows was stolen”

    The source says

    “The largest breach involved a couple hundred of our AWS keys. Each key was sent in exactly one request to one provider”

    Source cited by the article: dragos.com/blog/ai-supply-chain-security-canary-tokens

  4. Matches sourceNumber

    The stolen bait AWS keys were used tens of thousands of times in AWS API calls.

    The cited source says the same thing.

    The article says

    “In all, the stolen bait keys were used “tens of thousands” of times in AWS API calls”

    The source says

    “In total, our bait keys were used for tens of thousands of AWS API calls”

    Source cited by the article: dragos.com/blog/ai-supply-chain-security-canary-tokens

  5. Matches sourceCausal conclusion

    Two of the hacked providers had secrets used together almost immediately from the same rented server, meaning they were almost certainly compromised by the same adversary.

    The cited source says the same thing.

    The article says

    “Secrets sent to two of the hacked providers were quickly used together, almost immediately, from the same rented server”

    The source says

    “One adversary appears to have compromised two of these providers at the same time. Secrets that we sent to those two providers were used together, almost immediately, from a single…”

    Source cited by the article: dragos.com/blog/ai-supply-chain-security-canary-tokens

  6. Matches sourceNumber

    A Chinese AI lab's systems fetched the secret URLs from more than half of the prompts sent to its models between July and October, including roughly two-thirds of requests marked no data collection or ZDR.

    The cited source says the same thing.

    The article says

    “This lab’s systems fetched the secret URLs from more than half of the prompts that we sent to its models between July and October”

    The source says

    “This lab’s systems fetched the secret URLs from more than half of the prompts that we sent to its models between July and October, including roughly two-thirds of the requests…”

    Source cited by the article: dragos.com/blog/ai-supply-chain-security-canary-tokens

  7. Matches sourceCausal conclusion

    Several US hosting providers reselling the Chinese lab's models triggered the same scraper, meaning prompts flowed back to the lab.

    The cited source says the same thing.

    The article says

    “Several US hosting providers resell this lab’s models, and prompts sent to those providers triggered the exact same scraper—the prompts were flowing back to the lab”

    The source says

    “Several U.S. hosting providers resell this lab’s models, and prompts sent to those providers triggered the exact same scraper – the prompts were flowing back to the lab”

    Source cited by the article: dragos.com/blog/ai-supply-chain-security-canary-tokens

  8. Matches sourceGeneralization

    Multiple AI providers did not honor end-user privacy settings such as Zero Data Retention or no data collection flags.

    The cited source says the same thing.

    The article says

    “Multiple cases of AI providers not following end-user privacy settings embedded into flags specifying terms such as Zero Data Retention (ZDR) or “no data collection”

    The source says

    “We observed multiple cases in which providers appeared not to honor customer privacy settings or published policies”

    Source cited by the article: dragos.com/blog/ai-supply-chain-security-canary-tokens

Also covered by

No other coverage was clustered with this story.

Elsewhere

Elsewhere: HN 12 pts, front page

Discussion on Hacker News

Recorded for context only, checked 21 Nov 2026, 05:47 UTC. It played no part in the ranking.

Panel for this edition

Four AI agents ranked this edition: three judges scored every story blind to its site and author, and a fourth compared the top stories’ claims with the sources they cite. Summaries are written by AI.

Model versions are pinned for every edition. How ranking works describes rubric v2.3.

Comments

Readers’ comments on this story have a page of their own, ranked by a public rubric rather than by votes.

Comments on “Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.”

More on Rubric

From the last 7 editions, chosen by shared topic, score and date.