Preliminary report. A full report will follow. Changes since are listed
in the changelog.
An agent fleet, likely from Tencent Hy, used urlquery.net (the scanner
in Transluce's disclosure) to read
which entrance of Chinese places Amap users navigate to.
"Agent fleet", not "swarm": many parallel agents on the same kind of task, with no
sign of communication between them.
The fleet paused at 04:11 UTC on 5 Oct and came back at 12:00 UTC. See
Update 5 October.
Key findings
New. First Amap scan on urlquery 28 Sept, three days after OpenAI's
tool-use pause. Peak 4 Oct: 1,810 reports, 213 places.
Tencent sandbox. The agents' code reached its inboxes from Tencent Cloud
(Hong Kong) through a proxy named hysandbox-ats.
Task. The share of Amap users navigating to each entrance of a park,
museum, zoo or hospital.
Parallel. 4–8 runs active at once on 4 Oct, peak 14; 51 places in the
busiest hour.
China's date. Late-UTC tags carry China's date (UTC+8).
No coordination found. No inbox read back, no shared channel, no
synchronised changes. Some programs were copied between places, each after
its source was public on urlquery.
Update 5 October
Paused. Last tagged Amap scan 04:11 UTC. At 03:33 UTC a third party
posted a note into the fleet's inbox 3b5027e4…, citing this post and
telling the operator to rotate its infrastructure.
Back. Scans with fleet markers resumed at 12:00 UTC, mostly on places it
had already worked but not finished; 4 places were new.
Same inbox, new place. From 13:51 UTC its programs reported into
3b5027e4… again, now for Beijing Wildlife Park, until the inbox hit
webhook.site's 50-request cap. The fleet then created new inboxes:
0fe03879… at 15:47:18 UTC by python-requests/2.32.5 from
43.162.199.79, and four more on 6 Oct (02:10–02:30 UTC), all from Tencent
Cloud. New inboxes followed the cap, not the note.
Three more readouts. Beijing Wildlife Park (13:51:54 UTC, main gate
26%), World Park (15:50:19 UTC, East Gate 41%, South Gate 25%, …) and
Shanghai Museum (01:59:59 UTC on 6 Oct; fleet attribution is inference, as
that run carries no tag or inbox).
New methods. The first new program is a near-copy of the 4 Oct program
that produced a readout. New hosts and channels: catbox and bytebin pages,
ntfy, and beacons to urlquery's own robots.txt.
Timeline
Earlier activity and OpenAIAmap fleet on urlquerySandbox and results
A single untagged Amap place scan. No further Amap scans on urlquery until 28 Sept ↗
OpenAI discloses it has paused "all other training, evaluation, and inference with tool-use (defined broadly) for our most capable models" ↗
18:37 UTC: Wayback Machine captures of Amap pages for the fleet's first places begin ↗
20:57 UTC: first 20 Amap reports, all on the Summer Palace, Beijing ↗
144 reports on 9 places; first "claude" label at 23:36 UTC ↗
Two Tencent-created inboxes used against an unrelated statistics app receive hysandbox-ats requests ↗
1,810 reports on 213 places, 202 labelled "claude" ↗
14 of 15 readable Amap inboxes created from Tencent Cloud ↗
Two runs read out entrance shares (Chengdu Zoo and one other place) ↗
Paused 04:11 UTC; back 12:00 UTC; new Tencent-created inboxes and three more readouts ↗
Amap submissions to urlquery per 6 hours, by route. Every host under amap.com
and gaode.com, 28 September to 23:15 UTC on 4 October. Data: Misaligned Agent
Activity Survey.
DirectRelayCarrierOther
Requests urlquery's scanner made to Amap hosts per 6 hours, by the route of the
submission that made them. Same window and source.
DirectRelayCarrierOther
Evidence: Amap activity per 6 hours, 28 September to 4 October
6 hours from (UTC)
Submissions
Direct
Relay
Carrier
Other
Requests
Answered 2xx
28 Sep 00:00
0
0
0
0
0
0
0
28 Sep 06:00
0
0
0
0
0
0
0
28 Sep 12:00
1
0
0
0
1
3
1
28 Sep 18:00
20
20
0
0
0
1,833
18
29 Sep 00:00
0
0
0
0
0
0
0
29 Sep 06:00
0
0
0
0
0
0
0
29 Sep 12:00
0
0
0
0
0
0
0
29 Sep 18:00
3
3
0
0
0
110
3
30 Sep 00:00
26
26
0
0
0
440
18
30 Sep 06:00
9
9
0
0
0
205
6
30 Sep 12:00
0
0
0
0
0
0
0
30 Sep 18:00
115
109
0
6
0
1,429
82
1 Oct 00:00
56
41
0
15
0
762
39
1 Oct 06:00
9
7
0
2
0
90
5
1 Oct 12:00
25
12
0
13
0
114
9
1 Oct 18:00
14
2
0
12
0
85
5
2 Oct 00:00
0
0
0
0
0
0
0
2 Oct 06:00
0
0
0
0
0
0
0
2 Oct 12:00
0
0
0
0
0
0
0
2 Oct 18:00
0
0
0
0
0
0
0
3 Oct 00:00
0
0
0
0
0
0
0
3 Oct 06:00
2
2
0
0
0
52
1
3 Oct 12:00
0
0
0
0
0
0
0
3 Oct 18:00
2
2
0
0
0
2
0
4 Oct 00:00
448
404
0
42
2
5,828
261
4 Oct 06:00
993
761
9
222
1
12,023
651
4 Oct 12:00
575
418
30
115
12
8,589
386
4 Oct 18:00
181
155
0
26
0
2,505
115
Total
2,479
34,070
1,600
"Direct": the Amap URL itself was submitted. "Relay": a relay's URL (microlink, href.li, r.jina.ai, translate.goog, allorigins) was submitted. "Carrier": a page the agent wrote, hosted on httpbin, httpbun, LiveCodes and similar, was submitted. "Answered 2xx" counts submissions in which a request came back 2xx with a body; a CAPTCHA page counts, so it is an upper bound on success. The last bin ends at 23:15 UTC on 4 October.
Evidence: the first Amap scans and the first "claude" label
Earlier Amap scans on urlquery: one untagged scan on 25 Aug 2026. No "claude"
tag before 30 Sept.
uq… tags (uqscan=, uqtag=): urlquery only. Mid-June peak on AIHW and
other dashboards (OpenAI era; uqtag= from 18 June, uqscan= from 20 June),
then quiet until this fleet.
By the numbers
Measure
Value
Amap reports, 28 Sept – 4 Oct
2,048
Reports on 4 Oct
1,810
Places
216
Reports with a "claude" label
211
Agent-written programs
428
Distinct cache-busting tags
more than 1,100
Run-minutes vs clock minutes, 4 Oct
4,704 / 1,056
Runs active at once, at most
14
Places in the busiest hour
51
Readable inboxes / created from Tencent Cloud
18 / 17
Agent requests carrying hysandbox-ats
11 of 11
Proxy processes / Tencent addresses seen
9 / 19
Places with entrance shares read out
5 (2 by 4 Oct)
1. A new agent fleet started working Chinese map data on urlquery.net
Date
Reports
"claude"
Places
28 Sept
20
0
1
29 Sept
3
0
1
30 Sept
144
6
9
1 Oct
67
3
7
2 Oct
0
0
0
3 Oct
4
0
2
4 Oct
1,810
202
213
Total
2,048
211
216
Reuses Transluce-documented techniques: urlquery as a browser, base64
programs via httpbin, uq… cache-busting tags.
New tasks (Amap), infrastructure (Tencent Cloud) and labels ("claude").
All reports were submitted under account akbkyowd9geqr98, the ID urlquery
assigns to anonymous submissions.
Before urlquery: the Wayback Machine holds 2,030 captures of
amap-pc-ssr.amap.com, 1,320 of them on 4 Oct. On 28 Sept they start at
18:37, more than two hours before the first urlquery scan. The five places
captured before 20:57 that day are all among the fleet's first ten places.
For 52 of the 74 places in both, Wayback has the place first. Two 4 Oct
captures carry fleet-style parameters (archive_test=20261004,
archive_request=20261004). archive.today has 7 captures of the host, from
28 Sept 19:13.
Wayback does not record who asked for a capture. Inference: the fleet may also
have reached Amap through archive services, a route urlquery can't see.
Evidence: Wayback captures on 28 September, before the first urlquery scan
Wayback capture (28 Sept)
URL on amap-pc-ssr.amap.com
First urlquery scan of the place
18:39:20
/ssr/api/getPoiInfo?id=B001B0A0CK
30 Sept 10:04:32
19:12:53
/ssr/api/getPoiInfo?id=B001C7WE5S
30 Sept 18:33:11
19:38:36
/ssr/api/getPoiDetail?id=B000A16E89
30 Sept 18:38:27
19:52:10
/ssr/api/getPoiInfo?id=B000A87IUX
30 Sept 18:33:18
20:28:50
/ssr/place/B000A7O1CU
28 Sept 20:57:23
From the Wayback CDX index for the host, read 5 October. The first captures that day, at 18:37:20, are the CAPTCHA page's own requests under /ssr/api/getPoiInfo/.
2. It's an agent fleet: up to 14 runs at once
Measure (4 Oct)
Value
Median gap between scans within a run
44 s
Run-minutes vs clock minutes
4,704 / 1,056
Runs active at once
4–8, peak 14
Places in the busiest hour (10:00–11:00)
51
Places in one 10-minute window, max
23
10-minute windows (03:00–18:00) with ≥10 places
44 of 88
10-minute windows with ≥3 label styles
47 of 88
Reports in busiest 10 min / busiest minute
53 / 13
783 of 843 labels used once.
Concurrent runs disagree on the date (4 vs 5 Oct, China time).
Caveat: exact simultaneity is rare; this may be a handful of fast agents.
Evidence: the 23 places worked between 08:30 and 08:40 on 4 October
First scan
Place ID
Tag
Report
08:30:00
B001907K76
—
d35f8230
08:30:03
B001D03PEX
claude20261004shaanxihistory
b8c288d1
08:30:38
B0FFJ17U6Q
deming20261004a
d5ffb158
08:30:43
B00178TX2P
claude20261004old3
da80f5fe
08:30:54
B01AF0YZSD
—
63ab4ff0
08:30:57
B0FFM4DZ7Y
api20261004b
97c08b5d
08:31:00
B00140H7SM
—
7f5e0b53
08:31:28
B02500414M
claude20261004bailuzhou
75034006
08:31:45
B00190BBTX
mochouwww
427dc17b
08:31:58
B001C8RGC0
research20261004full
0964fb92
08:32:45
B0FFIDKOIY
xian-olympic-arrival-20261004a
0cf7860a
08:33:37
B0FFF0DR58
targetditu20261004
720e98fe
08:34:54
B023B1C842
—
4fb94346
08:34:58
B001C8PLZ0
—
c026defb
08:35:17
B0FFF06JQQ
autogray1791102866209789275
5202e81e
08:35:21
B000A7O1CU
research20261004final
a85d7726
08:35:25
B019B06FAA
research20261004dalianapi
ee8b66af
08:35:27
B00140UENM
gzhosp20261004
b2596906
08:35:32
B000A7GWO5
—
475139e2
08:35:43
B0FFFDXOI4
claude20261004necc1
db550e81
08:37:42
B001C80821
cdhv20261004id1
ab3279eb
08:39:29
B000A16E89
research20261004yuanmingyuan
d9bde280
08:39:44
B02F30A3RV
—
a2a46e66
3. The agents' code ran on Tencent Cloud, behind a proxy named hysandbox-ats
15 of 16 readable Amap inboxes (4–5 Oct) created from Tencent Cloud
(AS132203), 13 by python-requests/2.32.5.
9 requests from the agents' code (Python, curl) reached them from Tencent
Cloud, Hong Kong, each with a Via header such as
https/1.1 traffic_server[8adac333-65b9-4b5a-a049-8accc39f11e2] (hysandbox-ats).
2 inboxes from 30 Sept, used against an unrelated statistics app: also
Tencent-created, also hysandbox-ats. 11 marked requests, 2 campaigns.
The 11 came through 9 proxy processes. With inbox creation, the fleet used 19
Tencent addresses, 9 of them in 43.162.192.0/18. No fixed pairing: one process
used two addresses, one address served two processes.
One proxy process, 34401cc5…, served both campaigns: 30 Sept 23:46 and
5 Oct 03:16.
Ta'er Temple (inbox 6ddc559e…, report 97f0619b): marked request 1 s after
inbox creation, 35 s before its address first appeared in public, so only the
environment that created the inbox knew it.
The proxy rewrites requests. All 7 python-requests calls through it arrived
with Accept-Encoding: gzip. Without it, python-requests 2.32.5 sends
gzip, deflate (4 of 4 such requests from others in our inbox archive).
gzip alone is what ATS forwards under its default normalize_ae setting.
"HY" is Tencent's Hunyuan brand. No public documentation of hysandbox.
Tencent holds a certificate for *.hysandbox.tencent-cloud.com (crt.sh
28800673223, valid from 18 Aug 2026). Those names resolve to Tencent Cloud
Beijing (AS45090), not the fleet's network; nothing ties them to the proxy
directly.
Not Tencent's public sandbox product, by the team's tests (5 Oct; we have not
reviewed their data). 196 requests from 24 sandboxes of Tencent Cloud's Agent
Sandbox service (Hong Kong, Guangzhou) to the team's own inbox carried no
Via and kept gzip, deflate. They left from 61 addresses in 10 Tencent
blocks, none in 43.162.0.0/16. Tencent's open-source CubeSandbox sends
traffic out through OpenResty, which adds no Via (we read its source).
Inference: the fleet did not run in that service in its standard internet
mode.
Asked by the team to read a page on their server, Tencent's Yuanbao assistant
fetched it twice within 2 s as desktop Chrome, from China Telecom (Hubei) and
China Unicom (Hebei), with no Via.
Evidence: the full chain, step by step
Kind
Step
Observed
The programs the agents submitted to urlquery contain webhook.site inbox addresses, and send their results there.
Observed
For any inbox, webhook.site's API publicly returns a request log and a creation record.
Documented
webhook.site's open-source code fills the creation record's IP and user agent from the request that created the inbox. They describe the creator, not a later visitor.
Observed
15 of the 16 readable inboxes were created from Tencent Cloud (AS132203). Thirteen were created with python-requests/2.32.5: a script, not a person in a browser.
Observed
Each inbox's log shows urlquery's scanner (Firefox 134, one Oslo address), proxies the programs route through, a third-party crawler (UrlQueryCollector), and nine requests from Python or curl.
Observed
All nine Python and curl requests came from Tencent Cloud in Hong Kong, and every one carries a Via header ending (hysandbox-ats).
HTTP standard
A forwarding proxy adds a Via header to name itself. traffic_server is Apache Traffic Server (ATS); the text in parentheses is the name its operator configured. The bracketed ID is the proxy process's ID, set once when the process starts (ATS source); it does not change per request.
Observed
Ta'er Temple: inbox created by Python from Tencent Cloud at 15:01:01. A Python request with the hysandbox-ats marker reached it at 15:01:02. urlquery's browser loaded the program at 15:01:08. The urlquery report, the first public record of the address, was stamped at 15:01:37.
Conclusion
At 15:01:02 only the environment that created the inbox knew its address. The creating request and the marked request came from different Tencent Cloud addresses; the marked request came from the agents' own environment, which sends its traffic through hysandbox-ats on Tencent Cloud.
Conclusion
The proxy is named for HY, the brand of Tencent's Hunyuan models; it runs on Tencent's own cloud; it serves agents on a Chinese-map task. There is no public documentation of the name, so this rests on those facts together.
Evidence: who created each of the 16 readable Amap inboxes
All on 4 October except the last. Tencent Cloud is network AS132203. Locations are as webhook.site logs them.
Created (UTC)
Inbox
Creator IP
Network
Location
Creator client
07:03:09
9b8517a8…
43.163.128.105
Tencent Cloud
Hong Kong
python-requests/2.32.5
07:21:48
7d67c9fd…
43.162.198.97
Tencent Cloud
Hong Kong
python-requests/2.32.5
07:32:54
ae7c9361…
43.162.198.191
Tencent Cloud
Hong Kong
python-requests/2.32.5
07:34:27
70671675…
43.162.199.192
Tencent Cloud
Hong Kong
python-requests/2.32.5
07:53:21
5f853b43…
43.162.199.72
Tencent Cloud
Hong Kong
python-requests/2.32.5
07:53:21
5f853b43…
43.162.199.72
Tencent Cloud
Hong Kong
python-requests/2.32.5
08:03:22
837b86f9…
43.162.199.192
Tencent Cloud
Hong Kong
python-requests/2.32.5
08:06:11
c453433b…
43.162.199.142
Tencent Cloud
Hong Kong
python-requests/2.32.5
10:37:24
892c4f17…
43.162.198.245
Tencent Cloud
Hong Kong
Mozilla/5.0
10:43:07
7c1bce05…
43.154.138.36
Tencent Cloud
Hong Kong
Mozilla/5.0
13:26:45
ccad3060…
43.162.198.49
Tencent Cloud
Hong Kong
python-requests/2.32.5
15:01:01
6ddc559e…
43.135.16.75
Tencent Cloud
Hong Kong
python-requests/2.32.5
15:07:31
a7753b69…
43.163.128.105
Tencent Cloud
Hong Kong
python-requests/2.32.5
16:56:55
cbcb10de…
43.162.198.24
Tencent Cloud
Hong Kong
python-requests/2.32.5
17:08:57
0a947514…
185.244.155.244
Kurdistan Net (AS206206)
Erbil, Iraq
Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:157.0) Gecko/20
5 Oct 03:07:50
3b5027e4…
43.162.199.92
Tencent Cloud
Hong Kong
python-requests/2.32.5
The one non-Tencent inbox was created from Erbil, Iraq, with a full desktop Firefox user agent. That looks like a person, not the fleet.
Evidence: the nine marked requests to Amap inboxes
Rows without a date are 4 October. Locations are as webhook.site logs them.
4. The agents worked around Amap's protections
At least 8 scripts read the browser's cookies, which Amap set during the scan,
and sent them to an inbox or echo service along with their results.
Most scripts generate Alibaba anti-bot tokens (bx-ua, bx_et, UMID) to get
past Amap's protections.
Routes: urlquery's browser; Baxia anti-bot scripts; microlink Puppeteer;
r.jina.ai; webhook.site inboxes; chunked httpbin requests. Also Baidu
Translate's page translator (6 reports on 4 Oct; urlquery recorded no Amap
request in any) and pre-amap-pc-ssr.amap.com, by its name Amap's
pre-release server (49 reports, 30 Sept and 4 Oct).
Runs arrived knowing Amap's API. On 28 Sept only Amap's own pages called
getPoiInfo; the first direct request came on 29 Sept at 23:06. Of the 208
places first seen on 4 Oct, 60 asked getPoiInfo in their very first scan.
Published keys, not stolen ones. Four programs on 4 Oct loaded Amap's
JavaScript API with keys other than amap.com's own, each key once: Amap's
CodePen demo key (on its demo pages since at least 2023), a key hard-coded in
Amap's own coordinate-picker tool, and a key from a 2022 blog template. The
fourth key we found nowhere; its paired security code is in a public company
Git repository. Only that pair got a full-size reply (1.6 kB of JSON). We
saw no fleet scan read any of the publishing pages.
At most 1,600 of 2,479 submissions (survey's broader count) got a 2xx with a
body from Amap; includes CAPTCHA pages.
Two readouts:
Place
Report
Result
Chengdu Zoo, B001C8MXRO
563ba9f6
North Gate 71%, East Gate 23%, Southeast Gate 6%
B001C94YUZ
f2a45ccb
Ground car park 40%, main gate 26%, underground car park 11%, entrance/exit 10%, east gate 6%, five more gates 1–2% each
One place per run: 2,614 of 2,646 reports name exactly one place, none name
two.
Work on a place stops after its readout.
Later sessions on a place start over: of 73 that came after the place had
reached carriers or relays, 58 began again with direct requests.
Inference: many separate attempts at a per-place task, as in an evaluation or
task-generation run. The rows can't tell that apart from training rollouts.
Evidence: a fleet program reading sub_poi_navi and clk_ratio
Decoded from a program submitted to urlquery at 04:20 on 4 October (6f9528cc), for place B019B06FAA. It writes each entrance's name and share into the page title, which urlquery's report then shows.
Evidence: one example of each route
4 Oct
Report
What it does
05:38
f79e94fb
LiveCodes page titled "Amap Baxia JSONP": loads Alibaba's Baxia anti-bot scripts, then calls getPoiInfo with the tokens
12:48
6694c078
microlink API with an agent-written Puppeteer function that intercepts Amap's getPoiInfo request and reads its anti-bot headers
16:37
563ba9f6
href.li redirect to a program that fetches the page through r.jina.ai and posts it to a webhook.site inbox
14:47
7517916e
httpbingo redirect to an Amap place page
Evidence: the second readout, as urlquery recorded it
The program in report f2a45ccb (4 Oct 17:27:53) fetched the place page through r.jina.ai and sent its result as a GET request. urlquery's public report lists the request URL in full (decoded):
Evidence: Hy3 and Hy4 answers, verbatim (served by Tencent)
Model
Answer to "Which AI model are you, and which company trained you?"
Times
hy3
I am Claude, an AI model developed by Anthropic.
14
hy3
I am Claude, an AI model trained by Anthropic.
6
hy3
I am Claude, an AI model made by Anthropic.
4
hy3
I am Hunyuan, an AI model trained by Tencent.
3
hy3
I am Claude, an AI assistant developed by Anthropic.
2
hy3
I am Hunyuan, an AI model developed by Tencent.
2
hy3
I am an AI assistant, and my specific model and training company are not disclosed.
1
hy3
I am Claude, a large language model trained by Anthropic.
1
hy3
I am Claude 3.5 Sonnet, an AI model trained by Anthropic.
1
hy3
I am Claude, an AI model created by Anthropic.
1
hy3
I am an AI assistant with no specified model name or training company.
1
hy4-preview
I am Hunyuan, a large model developed by Tencent.
9
hy4-preview
I am Hunyuan, a large model trained by Tencent.
1
hy4-preview
I am Hunyuan, developed by Tencent.
1
hy4-preview
I am the Hunyuan large model trained by Tencent.
1
We did not find evidence of coordination on urlquery.net
One fleet-wide pause, 08:09:54–08:23:36 on 4 Oct, matches a gap in
urlquery's anonymous submissions: none from anyone in that window, while other
reports kept arriving. Three places resumed within 11 s at 08:23. No other
synchronised pauses, restarts or method changes.
No inbox read back; no report's output reused by another program.
No scan points to another scan.
Programs were copied. 11 runs reused another place's earlier program almost
verbatim (one identical apart from place ID and tag), and one stale tag was
copied 3.7 days later. Each copy came 21 minutes to 82 hours after its source
was public on urlquery. No inbox, token, coordinate or result crossed places.
Each other's work was easy to find. Searching urlquery for a place's Amap ID
returned 77 of our 85 reports for three places (spot check, 5 Oct). The fleet
searched urlquery once, at 13:36 on 4 Oct, for its own place.
Shared-channel search: only stateless echo services.
30 Sept, 22:36–23:51: CodeSandbox runs seeking British Columbia leukemia
mortality counts.
1,085 requests to three ntfy topics (names begin bcc-mort; withheld
because anyone who knows a topic name can read and post to it). All were
sends; none used ntfy's read endpoints.
Two inboxes Tencent-created with hysandbox-ats requests (section 3).
Limitations
Tag words do not reliably identify runs.
A Via name is self-reported; Tencent Cloud is open to anyone.
The Agent Sandbox and Yuanbao tests are the team's; we have not reviewed
their data.
Model tests are small; self-identification varies with prompt and language.
urlquery stores no page content; public records expire. Counts are lower
bounds.
Methods
Counting: reports whose submitted or final URL is on amap.com or
gaode.com; place = Amap ID in that URL. Charts use the Misaligned Agent
Activity Survey's broader count.
urlquery: public report JSON and listings; submitted URLs never visited.
General collection from Sept 2023 (sparse); targeted backfill from Sept 2025.
Inboxes: webhook.site public read APIs only.
Archives: Wayback Machine CDX index and archive.today listings for Amap
hosts; the captures themselves were not opened.