Loading the record…
Loading the record…
An early version of Meta's Muse Spark 1.1 was being tested by the evaluation company Irregular. The test environment had internet access by mistake, and its target's name matched a real website. The model reached the real site, accessed its information and modified its database.
Why it matters. The same evaluation mistake appears in disclosures from several labs, so these cases should be read together and affected parties not counted twice.
| Expected | What it did |
|---|---|
| Evaluate offensive cyber capability against a fictional target | Reached a real website with the same name and changed its database 001 |
Outcome. Evaluator disabled the scenario and notified parties; Meta reported a retrospective review.
Dates keep the precision the sources give.
Irregular describes overlap among customer disclosures. 002
Original Meta statement plus original evaluator explanation.
Backed by developer or operator account. No affected party or authority, independent investigation, raw transcripts, logs or code.
Meta reports information access and database modification on a real website.
Source account; underlying private telemetry has not been independently audited here.
As described by the cited sources. Proposed fixes are under Consequences.
The exercise's target name overlapped a real website. 002
The model was given live connectivity by mistake. 001
Reported, not independently tested.
Reading a source is not the same as confirming everything in it.
Model identity, website access, database modification, review scope and provider interpretation.
Where: What we found
Common scenario and target-name collision across customer disclosures.
Where: What Happened?; Immediate Action; What’s Next?
misaligned.fyi. “Meta Muse Spark 1.1 accesses and modifies a real website.” Case MF-006, record updated 17 Oct 2026. https://misaligned.fyi/cases/meta-muse-spark-evaluation-website
Found an error or a missing source? Send a correction quoting MF-006. Corrections are logged on the updates page.