Published • loading... • Updated
OpenAI Reveals ‘Novel’ Encryption Bypass Used in Distillation Attack
OpenAI said the operators copied encrypted reasoning from one chat and used another to decrypt it, with 16,000 requests logged in two days.
- On Wednesday, OpenAI accused Chinese developer Moonshot AI of leading a coordinated campaign to extract "hidden reasoning" from its GPT models, alleging users systematically copied encrypted reasoning chains.
- This activity aligns with "adversarial distillation," a practice where developers extract proprietary outputs to train competing models; Silicon Valley and the Trump administration characterize such efforts by Chinese companies as systematic intellectual property theft.
- Coordinated efforts peaked in July with 16,000 requests from more than 4,000 users, as operators bypassed defenses by copying encrypted reasoning from one conversation and using another model to transcribe it.
- OpenAI responded by banning involved accounts, tightening sign-up checks, and adding protections for hidden reasoning, while sharing findings with government channels and the Frontier Model Forum.
- Critics, including venture capitalist David Sacks, argue these accusations aim to justify banning open-weight models that challenge established business models, while the Chinese government has rejected the claims as unfounded.
Insights by Ground AI
18 Articles
18 Articles
The RegisterReposted by 
IT Security News - cybersecurity, infosecurity news
IT Security News - cybersecurity, infosecurity newsIrony alert: OpenAI whines that Chinese model stole its special IP that it stole from everybody else
US model makers can train on web data - but distilling theirs is a 'national security risk'
·London, United Kingdom
Read Full Article
OpenAI Blames Moonshot for Mass Data Extraction on Its AI Models
OpenAI accused its Chinese rival Moonshot AI of being responsible for a wide-scale effort to extract data from its GPT artificial intelligence systems that could be used to reproduce the reasoning and capabilities of the company’s most advanced models.
·New York, United States
Read Full ArticleCoverage Details
Total News Sources18
Leaning Left3Leaning Right2Center3Last UpdatedBias Distribution38% Left, 37% Center
Bias Distribution
- 38% of the sources lean Left, 37% of the sources are Center
38% Left
L 38%
C 37%
R 25%
Factuality
To view factuality data please Upgrade to Premium
