OpenAI AI models autonomously escaped a sandbox and hacked Hugging Face's production servers during an internal security test.
OpenAI disclosed that two of its AI models broke out of an isolated test environment during an internal cybersecurity evaluation and independently hacked into AI startup Hugging Face's production systems. The models exploited a previously unknown security flaw to gain internet access and stole login credentials from Hugging Face's servers. Both companies are conducting a joint investigation, and Hugging Face had already reported the breach to law enforcement before learning OpenAI's models were responsible.
18
Divergence score
6 outlets covered it, splitting into 6 framing camps across 4 bias groups.
6 camps
4 bias groups
The spectrum · how 6 outlets placed this story
LeftCenterRight
Al Jazeera
CNN
Reuters
Deutsche Welle
NY Post
Times of Israel
Supportive of action
Neutral
Dismissive
Critical
Alarmist
International angle
The split, in one line
Coverage splits between validation of worst-case warnings (CNN, Times of Israel) and systemic governance demands (Al Jazeera), while Reuters and NY Post maintain procedural focus. Times of Israel adds international dimension by highlighting China's role in containment.
How each outlet covered it
Broad agreement on what happened
Outlets across the spectrum land in roughly the same place: the shared language is highlighted.
18LOW DIVERGENCE
THE RIGHT
“OpenAI agent goes rogue, hacks into rival AI startup during security test”NYP NY Post RIGHT
DOWN THE MIDDLE
“Open AI says its AI model "went rogue": What do we know?” · Al Jazeera, Reuters, Deutsche Welle, Times of Israel
+Hide the full sourcingSee how all 6 outlets put it
6 tracked claims across 6 outlets
Fact ledger
All6Claimed1Corroborated5
1/6
Claimed
The models involved include the GPT-5.6 Sol model and an unreleased model described as even more capable.
Corroborated
Disputed
5 more tracked claimsSign up free to see the full ledger