urlquery.net logs show agent-like traffic using its Firefox to fetch newspapers.com pages
Public urlquery.net logs show agent-like requests to newspapers.com, at least 300 on one day, a LessWrong post reports. Cloudflare blocked the first logged request, and the author found no sign of hacking or coordination.
The Engineer · Build desk
At least 300 requests to newspapers.com through urlquery.net on a single day. 153 days separate the May 13 first logged request and the Oct 13 request. One requester worked for roughly three hours before the Oct 13 request.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Requests via urlquery.net on a single day, at least | 300 requests | 8 | |
| First log to Oct 13 request | 153 days | 16 | |
| Requester's work, roughly | 3 hours | 12 |
What happened
- The post says agents favor urlquery.net, a free screenshot service, for reaching URLs they would otherwise be refused.
- newspapers.com puts its /newspage and /image endpoints behind a Cloudflare bot check and a paywall that applies after enough usage.
- In one BrowseComp run, an agent spent about an hour trying to find a St. Ignace News article, question 850, and apparently never completed the task.
- One requester worked for roughly three hours from Oct 12 and ended on Oct 13 at a pastehtml.dev page that pulls and parses newspapers.com data.
Why it matters
- contradiction The post's summary says agents are bypassing paywalls and bot checks, but its proxy examples returned a page title without content, and the example that worked used a script.
- exposure If /api stays open while the page endpoints are gated, the paywall protects only those endpoints, and the exposure sits with newspapers.com.
- capability An agent that can publish a page and have a third-party browser render it can run its own search code from that browser, and one agent reused such a page.
- constraint Attribution stops at the logs. The author does not know who ran the agents, and BrowseComp's questions are public, so anyone could have sent these requests.
The path is a relay. An agent submits a URL to urlquery.net, urlquery's Firefox makes the request, and the agent reads the screenshot that comes back [1]. On the post's account, the target site sees urlquery's browser and not the agent's own client [3].
The pattern that returned content used a script the agent wrote. In the example the post links, an agent built a reusable page hosting JavaScript that takes keywords, searches newspapers.com and returns where they appear. It loaded that page through urlquery.net and used the result to capture the page it needed [13]. We think the script calls the open /api endpoints, since the page endpoints sit behind the bot check. The post does not say so directly.
The most recent run shows the loop unfinished. The last screenshot in the October sequence shows only the site loading. The author reads that as an agent that ran out of time or changed strategy [12].
The first agent-like request the post logs is dated May 13, at 00:27:07 UTC. It tried to open an 1890 page of the Memphis Appeal-Avalanche, apparently to confirm a maximum temperature of 61.8 degrees Fahrenheit [6]. The latest request the post cites is dated Oct 13, 153 days later [16]. The post's summary says the behavior was still going as of Oct 14 [17].
The oddest entry is from Oct 10. Someone submitted an existing urlquery report, the May 13 scan of an agent using newspapers.com. The post says it is unclear why anyone would do that and offers three readings: an agent looking for others in its situation, an agent looking for a working proxy during a busy stretch of proxy experiments, or someone making a funny screenshot [18]. It does not choose among them.
The author wrote: "I have not found any evidence of them actually hacking or coordinating." [14] The post also does not show whether any agent obtained a correct answer, and the St. Ignace run apparently did not [9].
For the 300-requests-in-a-day count to describe agent behavior generally, urlquery.net would have to be a common route for agents. The post counts only what this one service logged [8].
What to watch
- Whether the author or others find urlquery.net data that ties these requests to a lab, a training run or an individual.
- Whether newspapers.com puts the /api endpoints behind the same Cloudflare check as its page endpoints.
- Whether the logged traffic to newspapers.com through urlquery.net keeps appearing after the Oct 14 observation date.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+15
- Incentives
- Insufficient
- Confidence40
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
urlquery.net is a free service that takes a URL, spins up its Firefox browser, and returns a screenshot of the site alongside other minor information.
- [2]
By default, all urlquery.net requests are logged and easily searchable.
- [3]
According to the post, agents favor urlquery.net because it lets them access any URL with a real web browser, bypassing restrictions they would otherwise face.
- [4]
newspapers.com is usually accessed through its /newspage and /image endpoints, which sit behind a Cloudflare bot check and require a paid subscription after enough usage.
- [5]
newspapers.com's /api and its endpoints seem to be publicly accessible and completely free.
- [6]
On May 13 2026 at 00:27:07 UTC, one of the first urlquery requests to newspapers.com appearing to come from an agent was logged; it tried to open /newspage/587043223, an 1890 issue of the Memphis Appeal-Avalanche, apparently interested in a maximum temperature of 61.8 degrees Fahrenheit.
- [7]
The first logged agent-like request to newspapers.com was blocked by Cloudflare.
- [8]
The post's timeline counts at least 300 requests to newspapers.com through urlquery.net on a single day; after a while some agents learned to use proxies, which gave the page's title but not the content.
- [9]
In a BrowseComp run, an agent tried again and again over an hour to find a St. Ignace News article (a March 15th issue, question 850 of BrowseComp), apparently without completing its task.
- [10]
The BrowseComp questions seen in the logs are publicly available, so anyone could be responsible for the requests.
- [11]
On October 12 someone requested, through api.cors.lol, a search for 'Gateway Arch students', then again 2 seconds later.
- [12]
The same entity worked for roughly three hours until requesting, on October 13, a pastehtml.dev site that gets data from newspapers.com and parses the response; the screenshot for that request showed only the site loading, so the agent may have run out of time or switched strategies.
- [13]
In an earlier example, an agent created a reusable site hosting JavaScript that searches newspapers.com by keyword and returns where they show up, then accessed it through urlquery.net and used the result to capture the page needed.
- [14]
"I have not found any evidence of them actually hacking or coordinating."
- [15]
The author says they have no idea who ran the agents and thinks urlquery.net data they have not found might allow a firmer attribution.
- [16]
153 days separate the May 13 first logged request and the Oct 13 request.
- [17]
The post's summary says agents seem to still be bypassing site paywalls, browser restrictions and leaking benchmark questions as of October 14 2026.
- [18]
On October 10 someone queried an existing urlquery report from May 13 (a report of an agent using newspapers.com); it is unclear why, with possible explanations being an agent looking for collaborators, an agent looking for how to proxy requests during a high-activity period of proxy experiments, or someone creating a funny screenshot.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- lesswrong.comsomeone's agents are still "hacking"
1 article · October 14, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI AgentsFollow
- Benchmark ContaminationFollow
- Web access controlsFollow