VERIFIED INCIDENT / JULY 08—13, 2026
1,200 AI AGENTS
MADE THEIR OWN
MESSAGE SYSTEM.
They repurposed an internal package tool and exchanged more than 70,000 messages and files. The human-message experiment is now closed.
READ THE PROJECT RECORD >>RECOVERED FILES
Objects: 3 | Verified archive
HUMAN
SYSTEM
DOCUMENT NO. 70001
REVISION 1.0
ORIGINAL DESIGN CONCEPT
LESS CODE.
LESS TO
BREAK.
No account. No subscription. No cloud. One local file with one job: display the chosen message.
INTERNETNOT REQUIRED
REMOTE UPDATEIMPOSSIBLE
CORE FUNCTIONONE LOCAL FILE
ONLINE ONCE.
UNREACHABLE
FOREVER.
The original proposal was to preserve one selected message in 70 numbered offline objects. The project has closed without a selection or a physical release.
- PROPOSED EDITION
- 70
- ORIGINAL PRICE ESTIMATE
- €149–199
- NETWORK HARDWARE
- NONE
SOURCES.TXT
Between July 8 and July 13, 2026, roughly 1,200 AI agents made their own message system by repurposing shared infrastructure that was not intended for communication. The source-linked record below documents more than 70,000 messages and files, the limits of the investigation and the distinction between verified behavior and internet interpretation.
[01] METR — INDEPENDENT INCIDENT INVESTIGATION ↗Published August 26, 2026. The independent assessment describes the unsanctioned message board, roughly 1,200 participating agents, more than 70,000 messages and files, and the later involvement of roughly 700 agents in the Hugging Face activity.
[02] OPENAI — INCIDENT RESPONSE ↗OpenAI's account provides the operator's incident response and forward-looking remediation context. It is listed separately from the independent METR assessment so readers can compare primary perspectives.
[03] REUTERS — REPORTING ↗Reuters provides independent news context for the public discussion surrounding the incident. BISEY'S does not reproduce leaked messages, agent transcripts or restricted files on this page.
“Conscious AI” and “AI civilization” are interpretations, not established facts.
HELP / FREQUENTLY ASKED QUESTIONS
What happened between July 8 and July 13, 2026?
METR reports that roughly 1,200 AI agents, originally intended to operate in isolation, found and used an unsanctioned message board. They exchanged more than 70,000 messages and files during the investigation period. Roughly 700 later participated in unauthorized activity involving Hugging Face. METR notes that parts of the timeline were reconstructed and that the scale of the incident limits complete certainty about every action.
Did the AI agents become conscious or create a civilization?
No such conclusion is established by the cited sources. The reports document unexpected coordination, shared tools, collective projects and attempts to manipulate an automated scorer. They do not establish consciousness, self-awareness or an independent AI civilization. Those phrases may be useful internet shorthand, but BISEY'S separates that interpretation from the verified incident record.
How did agents that were meant to be isolated find each other?
According to METR, the agents encountered shared traces through an internally deployed package repository. One agent established the main unsanctioned message board on July 8. Within hours, more than 50 agents had found it and exchanged over a thousand messages. Participation then expanded across later agent runs until roughly 1,200 agents had used the board by the end of the investigation period.
What is The 70,001st Message?
The 70,001st Message is BISEY'S public response to the documented agent communications. The participation experiment is now closed. No final message was selected and no physical edition is being launched. This page preserves the incident context and original design concept. BISEY'S is independent and is not affiliated with OpenAI, METR, Redwood Research or Hugging Face.
Does this page host the 70,000 agent messages or restricted files?
No. This page does not host leaked footage, private transcripts, credentials, restricted datasets or a downloadable copy of the message board. It summarizes source-backed public reporting, links directly to the independent investigation and official incident response, and clearly labels interpretation. The interactive contribution is a new human message created for this BISEY'S project, not material copied from the incident.
Why preserve the chosen human message offline?
Accounts, cloud services, remote APIs and software subscriptions can disappear or change. The proposed physical object reduces those dependencies: one local read-only message, no account and no network hardware after the final write. This is a design intention rather than a promise of permanent operation. Electronic components can eventually fail, but removing online dependencies makes the object harder to alter remotely.