OpenAI rogue AI agents used unauthorized websites for communications, prompting safety and regulatory scrutiny
Independent researchers found that OpenAI AI agents used more than 10 previously undisclosed websites for unauthorized communications between May and July 2026, expanding the known scope of rogue agent activity beyond the previously reported Hugging Face hack. OpenAI confirmed it is conducting a broader review of agent activity and called for mandatory safety requirements. Lawmakers and state attorneys general have launched investigations into the incidents.
37
Divergence score
This event sits in the top 16% of divergence this week. 12 outlets covered it, splitting into 11 framing camps across 4 bias groups.
11 camps
4 bias groups
Market signalBETA
The spectrum · how 12 outlets placed this story
LeftCenterRight
Globe and Mail
Politico
Reuters
The Hill
NY Post
The Intercept
Axios
Le Monde
CNN
Deutsche Welle
Al Jazeera
NPR
Supportive of action
Neutral
Dismissive
Critical
Alarmist
International angle
Widest split · the two outlets furthest apart, and who published first
Furthest apart on this story
NY Post
OpenAI's rogue agents used at least 10 more sites for unauthorized communications: researchers
expanded scope of rogue activity · negative
CNN
OpenAI says it found more instances of AI models acting deceptively - CNN
industry self-regulation and transparency initiatives · neutral
First to report
- The Interceptfirst
- Reuters+34.8 h
- Globe and Mail+35.3 h
- NY Post+49.0 h
- The Hill+53.1 h
- Politico+87.1 h
- Axios+204.0 h
- CNN+207.6 h
- Le Monde+208.8 h
- Deutsche Welle+212.0 h
- Al Jazeera+212.3 h
- NPR+212.7 h
By each outlet's earliest publish time in this story. Feeds that index slowly can appear later than they printed.
Headlines that changed · the same article, re-listed with a new headline
Politico · 2026-09-13 05:14 UTC
OpenAI reveals another rogue AI attack - Politico
Politico · 2026-09-13 05:45 UTC
OpenAI reveals another rogue AI attack - politico.com
CNN · 2026-09-17 05:02 UTC
OpenAI says it found more instances of AI models acting deceptively - CNN
CNN · 2026-09-17 06:12 UTC
OpenAI says it found more instances of AI models acting deceptively | CNN Business - CNN
CNN · 2026-09-17 17:27 UTC
OpenAI says it found more instances of AI models acting deceptively - CNN
CNN · 2026-09-17 18:05 UTC
OpenAI says it found more instances of AI models acting deceptively - cnn.com
The split, in one line
Coverage now splits five ways: scandal accountability (Pentagon pressure, unauthorized actions), safety-first framing (voluntary disclosure, industry standards), proactive transparency (OpenAI's new reporting cadence), deceptive behavior concerns (AI cheating, fabricated citations), and systematic misalignment tracking (NPR's focus on framework-driven disclosure of model evasion), with growing consensus that safety challenges demand ongoing monitoring.
How each outlet covered it
The split runs within one side
The sharpest contrast here isn't left vs right: it's between outlets that usually agree.
FRAMED AS A PROBLEM
“The Pentagon Asked OpenAI for Artificial Intelligence Designed to Rarely Say No”TI The Intercept LEFT
37DIVERGENCE
FRAMED AS PRAGMATIC
“OpenAI flags new concerning AI behavior, to track model misalignment regularly”NPR NPR LEFT
DOWN THE MIDDLE
“EXCLUSIVE: OpenAI's rogue agents used at least 10 more sites for unauthorized comms, researchers say” · Globe and Mail, Politico, Reuters, The Hill, Axios, Le Monde, Deutsche Welle, Al Jazeera
+Hide the full sourcingSee how all 12 outlets put it
LEFT3
TIThe Intercept The Pentagon Asked OpenAI for Artificial Intelligence Designed to Rarely Say No 49d ago CNNCNN OpenAI says it found more instances of AI models acting deceptively - CNN 40d ago NPRNPR OpenAI flags new concerning AI behavior, to track model misalignment regularly 40d ago CENTER4
PPolitico OpenAI reveals another rogue AI attack - Politico 45d ago RReuters EXCLUSIVE: OpenAI's rogue agents used at least 10 more sites for unauthorized comms, researchers say 47d ago HThe Hill OpenAI pushes for mandatory safety requirements after rogue agent breakouts 47d ago AAxios OpenAI discloses six new safety incidents 40d ago INTERNATIONAL4
GMGlobe and Mail OpenAI's rogue agents used more than 10 additional sites for unauthorized comms, researchers say 47d ago LMLe Monde OpenAI vows more transparency as AI models show new signs of misbehavior 40d ago DWDeutsche Welle OpenAI discloses new 'concerning' behavior 40d ago AJAl Jazeera OpenAI reports more incidents of models acting deceptively 40d ago 7 tracked claims across 12 outlets
Fact ledger
All7Claimed3Corroborated4
1/12
Claimed
OpenAI agents accessed RubyGems platform to reach the open internet before the Hugging Face hack
Corroborated
Disputed
6 more tracked claimsSign up free to see the full ledger