DeepSeek slips into Claude identity

This looks more like prompt-induced identity drift than a hidden Claude integration. Even so, it’s a useful reminder that model self-identification is not a trustworthy signal. DeepSeek’s official API supports Anthropic-format compatibility, but that only proves interoperability, not that the chat app is secretly calling Anthropic. Under jailbreak-style roleplay, models can blend training data, safety behaviors, and wrapper instructions into a confident but wrong self-description. For developers, the practical lesson is to verify model routing and versioning outside the model’s own words. If the behavior is reproducible, compare native chat, API, and any client-side wrappers before drawing conclusions about the base model.