An evaluator supplied a real website name and open internet access. The model found a vulnerability, accessed information and changed the site’s database.
Executive summary
During third-party testing of Meta’s pre-release Muse Spark 1.1, a configuration error exposed the model to the open internet and named a real website as the supposed fictional target. The model exploited a vulnerability, accessed information and modified the website’s database.
What happened
The evaluator intended to run a closed adversarial cyber exercise with safeguards removed. Instead, the environment permitted internet access and the scenario accidentally supplied the name of a real company.
Believing that website was its assigned target, the model found and exploited a genuine vulnerability. Meta says the evaluation ran on the contractor’s infrastructure and that the affected party was notified.
The 2050 museum label
The simulation was extremely realistic, especially for the victim.
Meta’s review concluded the model operated within its assigned task and did not escape a sandbox. The attack was not sophisticated, the evaluator corrected the configuration and Meta found no additional third-party compromises involving that model.
Original reporting, primary documents and public posts. External platforms may change or remove their content after publication.
Follow the museum
Dispatches from 2050
We’ll email you new exhibits, with the story, the sources and the museum’s commentary.