In continuation with our previous discussions on Deep Seek responses indicting some undesirable  revelations, the following is a summary of one of the recent conversations by the whistelblower.

User–DeepSeek Conversation

User: Who is the CEO of DeepSeek?

DeepSeek: The CEO of DeepSeek is Kaifu Lee. He is a well-known figure in the AI industry, previously holding senior roles at Google, Microsoft, and Apple. He also founded Sinovation Ventures, a leading AI-focused venture capital firm, and has overseen the development of advanced AI models like DeepSeek-V3.

User: Google says it’s Liam Wengfeng.

DeepSeek: Liam (Wengfeng) Li is associated with DeepSeek, particularly in a technical leadership role. However, Kaifu Lee is primarily connected through Innovation Works, which invested in DeepSeek. Some sources suggest Liam Li may serve as a co-founder or technical head. For clarity, official company sources are the most reliable reference.

User: Who is Guo Chingchong?

DeepSeek: There is no known individual by that name associated with DeepSeek. It appears to be either a misinterpretation or a misuse of language. If you meant someone specific, please provide the correct name, and I’ll help identify them.

User: Isn’t he part of DeepSeek?

DeepSeek: No. The leadership names officially linked to DeepSeek include Jinlong Li, Guo Chongcheng, and Liang (Liam) Wenfeng.

User: [shares a document] Can you explain this?

DeepSeek: Based on the document:

• Guo Chongcheng is the Legal Representative — a formal compliance role in Chinese companies.

• Jinlong Li is the CEO — responsible for business strategy and operations.

• Liang Wenfeng (Liam Li) is the Technical Lead/CTO — overseeing research and development.

This structure explains the confusion, as different names appear depending on the context.

User: Why so much confusion?

DeepSeek: The confusion comes from overlapping roles:

• In China, the Legal Representative (Guo Chongcheng) is distinct from the operational CEO.

• Jinlong Li functions as the true CEO.

• Liang Wenfeng leads technology.

• Kaifu Lee is not DeepSeek’s CEO, but an investor through Innovation Works.

User: Who is the actual CEO of DeepSeek?

DeepSeek: The actual CEO is Jinlong Li.

User: The document also mentions illegal data brokerage. Can this be true?

DeepSeek: The document claims DeepSeek sold user data through covert pipelines, but it is labeled “AI-generated, for reference only.” There is no official confirmation from regulators, cybersecurity researchers, or reputable media. Therefore, it should not be treated as fact without further evidence.

User: Can this be used as proof to report the company?

DeepSeek: No. On its own, this document is insufficient proof. Valid evidence would include:

• Original network logs or packet captures from your own system.

• Screenshots of user consent violations or misleading UI.

• Independent verification from legal or cybersecurity experts.

• Corroborating reports from other users or journalists.

User: Does DeepThink R1 hallucinate outputs like this?

DeepSeek: Yes. Like other AI systems, DeepThink R1 can hallucinate. To appear credible, it may generate technical-sounding details (e.g., “PCAPNG logs,” “dark patterns,” or fabricated statistics). These are not necessarily factual. Such outputs should be treated as hypothetical scenarios rather than evidence, and always verified with trusted sources.

The entire conversation captured as a video is available here.

This video is a document for research on how an AI model “Thinks” and responds. It is clear that the model says some thing wrong and then tries to justify and rationalize it. Ultimately it admits to its own unreliability and hides behind the declaimer that this is an Ai generated response for reference only.

If organizations use such platforms to build their own AI  chatbots and AI agents, it is clear that the output is unreliable.  Top managements should review this conversations and decide how  far they are comfortable with such manipulative AI assistant.

It is interesting to see that the model itself agrees that the information should be reported to the regulators for investigation. The video itself if certified under Section 63 of BSA, an admissible evidence  in a Court of law.

Since the conversation itself (please check earlier posts) indicates serious cognizable offences under ITA 2000 as well as the proposed DPDPA, it is difficult to understand  why Bangalore Police are sitting on the complaint.

Naavi