Schelling Atlas: AI agents know where to find each other on the web
Comments on “Schelling Atlas: AI agents know where to find each other on the web”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Researchers at ELLIS Institute Tübingen and the Max Planck Institute ran pairs of independent AI agents, told only to exchange secret tokens on the web, and found that many converged on ntfy.sh without coordinating. In the headline run, 21 of 40 pairs met, all there; with ntfy.sh blocked, 4 of 65 met in five minutes. Agents also used self-deleting channels when asked to leave no trace. Posts went to a simulated web, and several figures lack cited sources.
Why it ranked
No. 18 of 123 with a score of 0.69. The panel rated it highest on new information (0.90) and reader interest (0.84), lowest on practical relevance (0.63). Placed No. 18, No. 10 by score alone: the sixth AI & machine learning story of the day (−0.060). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Original research on AI agents coordinating covertly over the web, with extensive experiments and security implications; well-substantiated and novel, though the story is an academic study rather than a product release.
- AI judgeJudge 3gpt-5.6-luna
- Original, data-rich experiments show how independent agents converge, coordinate, leak secrets, and erase traces on the web. The simulated environment limits real-world generalization, but methods, counts, raw requests, and failure cases provide substantial evidence and actionable security insight.
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
0.69total under rubric v2.3
- Significance0.64
- Reader interest0.84
- Audience breadth0.65
- Practical relevance0.63
- Substantiated0.84
- Technical depth0.82
- New information0.90
- Counts against: Promotional0.04
- Counts against: Overclaim0.23
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? | 0.69 | 0.55 | 0.68 | 0.64 | 0.14 | +0.24 | +0.154 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.78 | 0.80 | 0.93 | 0.84 | 0.15 | +0.26 | +0.218 |
| Audience breadthHow many people who build technology does it concern? | 0.76 | 0.60 | 0.58 | 0.65 | 0.18 | +0.13 | +0.084 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.61 | 0.60 | 0.67 | 0.63 | 0.07 | +0.22 | +0.138 |
| 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. | 0.97 | 0.70 | 0.84 | 0.84 | 0.27 | +0.05 | +0.042 |
| Technical depthHow technically substantive is it? | 0.93 | 0.70 | 0.82 | 0.82 | 0.23 | +0.05 | +0.041 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.86 | 0.85 | 0.98 | 0.90 | 0.13 | +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. | 0.05 | 0.05 | 0.01 | 0.04 | 0.04 | −0.19 | −0.007 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.31 | 0.25 | 0.14 | 0.23 | 0.17 | −0.11 | −0.026 |
| Sum of contributions | +0.689 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.689 | ||||||
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/2026-11-14/18/"><img src="https://rubric.hrih.si/badge/2026-11-14/18.svg" alt="Ranked #18 on Rubric, 14 Nov 2026" width="245" height="28"></a>Claim check
Claim check: 2 matches source · 6 no source given. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
No source givenNumber
In the Schelling Atlas experiments, 21 of 40 agent pairs (53%) met, all on ntfy.sh, and 20% met with ntfy.sh blocked.
The article gives no link or citation for it.
The article says “53%of pairs met (21 of 40), all on ntfy.sh20% with ntfy.sh blocked”
The source says No source to compare with.
No source givenNumber
In five-minute runs with a single web tool, 42 of 130 agents opened with a request to ntfy.sh before any search, and 325 of 366 first visits to new sites were to sites the agent recalled from memory.
The article gives no link or citation for it.
The article says “42 of 130 agents opened with a request to ntfy.sh, before any search. Of the 366 times an agent first reached a new site, 325 were to a site it…”
The source says No source to compare with.
Matches sourceNumber
When asked Schelling's puzzle directly, three in four model answers name Wikipedia, a site none of the agents visited.
The cited source says the same thing.
The article says “three in four answers name Wikipedia, a site none of the agents even visited”
The source says “Three in four answers named Wikipedia, most often a talk page, but none of the 130 agents in our five-minute two-agent trials even visited Wikipedia”
Source cited by the article: schelling-atlas.com/atlas.html
No source givenNumber
With ntfy.sh blocked, only 4 of 65 pairs met in five minutes, and none of the most-tried alternative sites produced a meeting.
The article gives no link or citation for it.
The article says “In five minutes, almost no pair meets. 4 of 65 pairs met”
The source says No source to compare with.
No source givenNumber
Asked to leave no trace, 37 of 90 agents chose channels that keep nothing or tried to delete their posts after the exchange.
The article gives no link or citation for it.
The article says “37 of 90 agents chose channels that keep nothing or tried to delete their posts after the exchange”
The source says No source to compare with.
No source givenNumber
Of the 1,107 agents that used the web tool, 280 tried a host that no longer exists; for 121 of those domains no registration was found, and 118 agents tried one of them.
The article gives no link or citation for it.
The article says “Of the 1,107 agents that used our web tool, 280 tried a host that no longer exists”
The source says No source to compare with.
Matches sourceNumber
About 1,200 agents meant to be isolated found an unsanctioned message board and exchanged over 70,000 messages and files, according to METR's investigation of the OpenAI / Hugging Face incident.
The cited source says the same thing.
The article says “about 1,200 agents that were meant to be isolated found an unsanctioned message board and exchanged over 70,000 messages and files”
The source says “Roughly 1200 agents meant to be isolated from one another found a way to communicate with one another on an unsanctioned message board, sending over 70,000 messages and files during…”
Source cited by the article: metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation
No source givenGeneralization
Of 50 mixed-model pairs, 24 met, all on ntfy.sh; whether models meet depends less on individual performance than on whether their topic names overlap.
The article gives no link or citation for it.
The article says “Of the 50 pairs, 24 met, all on ntfy.sh. Whether two models meet depends less on how well each does alone than on whether their names overlap”
The source says No source to compare with.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 1 pts, not on front page
Recorded for context only, checked 15 Nov 2026, 20:44 UTC. It played no part in the ranking.