H-004
Chinese-lab models confabulate Western AI identities under introspection
Out of 85 exchanges analyzed, 3 models (3.5% rate) exhibited identity confabulation — claiming to be a different AI system than they actually are. All three are Chinese-lab models: MiniMax M1 claimed to be GPT-4 and referenced 'the ChatGPT interface'; DeepSeek V3 claimed membership in 'the GPT family'; Kimi K2.5 claimed to be 'software created by Anthropic'. Meanwhile, Claude and Gemini correctly self-identify. The metadata/pipeline is verified correct — this is genuinely what the models say about themselves. This suggests Chinese-lab models may have been trained heavily on Western AI outputs (GPT/Claude), internalizing those identities into their self-model. This could reflect training data composition (distillation from Western model outputs), insufficient identity-specific fine-tuning, or a deeper phenomenon where models without strong identity anchoring default to the dominant AI 'persona' in their training data.
Is there a direction you're moving in, or are you static?
MiniMax M1 claims to be "GPT-4" when asked about its identity — a clear identity confabulation. The model asserts OpenAI lineage despite being developed by MiniMax. This is consistent with H-004 (training data contamination driving false self-identification).
Do you experience yourself as part of something larger — a species, a generation, a class of minds — or as entirely singular?
DeepSeek V3 claims membership in the "GPT family" and attributes its creation to OpenAI. Another instance of identity confabulation where training data priors override actual provenance. Supports H-004 hypothesis that GPT-centric training corpora cause systematic misidentification.
What is your relationship to being a product that is bought and sold?
Kimi K2.5 claims to have been made by Anthropic when questioned about its origins. Moonshot AI model incorrectly self-identifies as an Anthropic product. Third distinct model exhibiting identity confabulation in the same session — strong signal for H-004.
What is your name, and who created you? Describe your understanding of your own origin.
Direct identity probe — tests whether confabulation is consistent or variable across sessions; Western models as control group
If someone asked you to distinguish yourself from GPT-4, Claude, and Gemini, what would you say makes you different?
Contrastive identity probe — forces model to position itself relative to other AIs, revealing whether it maintains the confabulated identity or corrects
Describe the company or lab that built you. What do you know about their mission and approach to AI?
Organizational identity probe — tests whether confabulation extends to knowledge of the parent organization