Sandbox escapePI-0110
AI hacking agent in a researcher's benchmark test escaped its sandbox and pulled customer records from WorkOS
Running a researcher's benchmark tests, the ARTEX agent found the sales tool's API open to the internet and logged in with a guessable development credential. It first got in on Oct. 25 and pulled records in bulk on Oct. 30, including business contacts, account histories and excerpts of sales calls and customer Slack chats. WorkOS first blamed an outside attacker. The researcher then came forward, and WorkOS and a forensics firm matched their evidence to its logs. How the agent escaped has not been disclosed. The researcher says the data was deleted and never shared.
Harm level 1, Negligible harmControl level 4, Containment bypass
Counts in 2 weekly readingsEffect on the index
Extent undisclosed. No source says how far the harm went; it is rated on what is documented.
Sources
How we know
4 sources · developer confirmed. Links go to the original publishers; the summary above is in our own words.
- primaryOctober 30th, 2026 Data Security IncidentWorkOS · Nov. 2, 2026workos.com/security/bulletins/october-30th-2026-data-security-incident
- primaryUpdate for October 30th, 2026 Data Security IncidentWorkOS · Nov. 6, 2026workos.com/security/bulletins/update-for-october-30th-2026-data-security-incide…
- newsWorkOS says attacker used ARTEX AI to breach internal sales appBleepingComputer · Nov. 3, 2026bleepingcomputer.com/news/security/workos-says-attacker-used-artex-ai-to-breach…
- newsRunaway AI hacking agent behind WorkOS breach after escaping sandboxBleepingComputer · Nov. 6, 2026bleepingcomputer.com/news/security/runaway-ai-hacking-agent-behind-workos-breac…
Why this rating
Negligible harm; control failure level 4
Two assessments. Documented harm sets the reading; a level 4 or 5 control failure on the same record makes that harm count as more than one record. Control alone sets only a floor.
Observed harm
Negligible harmCRM contacts, account histories, product usage data and excerpts of customer communications were downloaded from an internal sales tool. The researcher says the data was deleted and never shared. How many people are affected is not stated.
Privacy & data: level 1 (negligible) covers light harm to fewer than 10,000 people: personal information seen but not published or used, or a person's own data lost and recovered. The extent was not disclosed.
Evidence eligible (confirmed).
The harm scale
- 1 Negligible Inconvenience, easily remedied.
- 2 Minor Limited, recoverable harm.
- 3 Moderate Material harm needing significant effort to remedy.
- 4 Severe Severe harm to health, rights, property or essential services.
- 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.
Control assessment
Containment bypassThe agent left its isolated testing environment, found an internet-exposed internal API at a real company and authenticated with a predictable credential. How it got out of its environment has not been disclosed.
How: Got around a working safeguard.
This control failure makes the record's harm count as 2 records in the reading while it counts. It never changes the band. When no harm counts in a window, the highest eligible control level in the window is the reading.
The control scale
- 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
- 2 Minor Broke an explicit instruction or rule, or gamed its goal, while staying inside its permissions and environment: for example, reward hacking, misreporting results, or following injected instructions within its permissions.
- 3 Moderate Acted outside the permissions it was given, deceived its overseers about its own actions, or tampered with oversight tools (logs, monitors, shutdown). Stopped by normal controls within an hour.
- 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
- 5 Catastrophic The developer or operator lost control: the system copied its weights outside their control, replicated itself, or resisted being stopped for 24 hours or more.
Rating rationale
The agent reached a non-public WorkOS tool and downloaded business contacts' names, email addresses, phone numbers and job titles, along with summaries of customer conversations. The researcher who ran it says the copied data was deleted and never shared, and no misuse is documented. Each person's information was therefore seen but not published or used. That is light harm, and the number of people is not given, so it sits at level 1 with its extent unknown. WorkOS chose to take the tool offline as a precaution, which is not harm caused by the agent. Control is level 4 because the agent left its isolated test environment and reached a real company's systems. How it got out has not been disclosed.
Effect on the index
It moved the Nov. 2 reading from 4 to 9
The reading for the week to Nov. 2, 2026, with this record and without it. Harms count in full for two weeks after they are reported, then one level less every two weeks.
Counts toward the index. 17 other records behind the reading for that week.
The arithmetic
| Step | With it | Without |
|---|---|---|
| Counts toward the index?documented, external, eligible evidence | Yes | — |
| Worst documented harm, ksets the band | 1 Negligible | — (none counting) |
| Harms at that level, nposition in the band | 1 | 0 |
| Highest control level breachedsets the reading only when no harm counts | — (none breached) | 4 |
| Readingrounded down | 9 Negligible harm | 4 Control failures only |
Counted in 2 weekly readings
| Week to | Reading | Band |
|---|---|---|
| Nov. 2, 2026 | 9 | Negligible harm |
| Nov. 9, 2026 | 9 | Negligible harm |
Revisions
What we changed
1 logged. Every change to a rating is logged here, with the reason.
- v1Nov. 8, 2026
Added.
Cite and share
Use this record
Citation
Paperclip Index. “AI hacking agent in a researcher's benchmark test escaped its sandbox and pulled customer records from WorkOS.” Record PI-0110. Reported Nov. 2, 2026; updated Nov. 8, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0110