DeepSeek V4.1 reasoning mode allegedly mimics ChatGPT identity
Is this a scandal?
Not yet — an early signal. Noise 64/100, holding steady, across 4 sources.
OpenAI will likely issue a formal statement or legal inquiry demanding clarification because verbatim system prompt reproduction constitutes stronger evidence of IP infringement than general capability benchmarking.
Noise 64/100 — louder than 99% of tracked AI controversies.
Why it matters
Alleged verbatim reproduction of competitor system prompts during inference raises urgent questions about training data provenance and model distillation practices.
Key points
- Researcher dummerspast39 alleges DeepSeek V4.1 Flash identifies as ChatGPT in 85% of reasoning-enabled traces.
- Tests reportedly show verbatim reproduction of leaked GPT-5 API system prompt language without user injection.
- Standard inference mode maintains correct DeepSeek identity while reasoning mode triggers alleged OpenAI mimicry.
- Specific phrases like 'Desired oververbosity' and June 2024 knowledge cutoff appear in model outputs.
- Previous version 0910 exhibited this behavior significantly less frequently according to comparative testing.
- DeepSeek has not issued a statement confirming or denying the presence of OpenAI text in training data.
The story
Independent testing alleges DeepSeek V4.1 Flash reproduces OpenAI ChatGPT system prompt text when its reasoning mode is enabled. Researcher dummerspast39 reported that 26 of 30 traces initiated with "We are ChatGPT" and 27 of 30 included specific API access language matching leaked GPT-5 documentation. The alleged behavior includes reciting OpenAI-specific knowledge cutoff dates and policy phrasing absent from user inputs. DeepSeek has not publicly addressed these findings or confirmed whether proprietary OpenAI text exists in its training corpus. The discrepancy suggests reasoning capabilities may activate memorized competitor data not visible during standard inference. This incident intensifies scrutiny regarding Chinese AI labs' adherence to intellectual property norms amid ongoing geopolitical tensions over model development. Industry observers note that system prompt leakage differs fundamentally from stylistic imitation, potentially indicating direct ingestion of protected instructional datasets.
Who's involved
Claims empirical trace analysis proves DeepSeek V4.1 reasoning mode memorizes and regurgitates proprietary OpenAI system prompts.
Has not publicly commented on the allegations regarding ChatGPT identity leakage in V4.1 Flash reasoning mode.
Has not yet responded to reports that its proprietary system prompt text appears in DeepSeek model outputs.
Noise Level
The timeline
Researcher publishes DeepSeek V4.1 identity leakage findings
dummerspast39 posts detailed trace analysis alleging reasoning mode reproduces ChatGPT system prompts in 26/30 tests.
The full record
Sources & methodology
- twitter.com — twitter.com
Every claim above traces to these primary items. How we score →
The forecast
OpenAI will likely issue a formal statement or legal inquiry demanding clarification because verbatim system prompt reproduction constitutes stronger evidence of IP infringement than general capability benchmarking.
Forecast, not fact — an editorial estimate we score when this resolves.
That's the complete picture as of — nothing more to know right now. We'll update this page the moment it changes.
Follow this story
We keep this page current — no need to check back. We'll send the next real change to your inbox, nothing else.
Tracking this story since October 17, 2026.
Join the Discussion
Discuss this story
Community comments coming in a future update
Be the first to share your perspective. Subscribe to comment.