The AI landscape is rapidly evolving, with new models vying for attention and market share. One of the latest entrants, DeepSeek, has generated significant positive buzz, but there are compelling reasons to approach this model with skepticism. Before we go into details, we would like to remind Carleton faculty, staff, and students that you have access to Google Gemini under an agreement that ensures that our prompts will not be used as training data. While some users may be intrigued by DeepSeek’s purported capabilities, several red flags suggest that it may not be a trustworthy or desirable alternative to other large language models (LLMs). Below are some of the most pressing concerns.
1. Privacy and Data Protection Concerns
One of the most critical issues with DeepSeek is its policy on data collection and usage. Unlike some AI providers that offer options for user privacy, DeepSeek’s terms of service clearly state that it will collect and utilize user interactions. This raises concerns about how this data is stored, who has access to it, and whether it could be shared with third parties, including governmental entities. Given the increasing importance of data security, users should be aware that engaging with DeepSeek means their data may be retained and used in ways that are not fully disclosed. Unlike other AI companies that explicitly outline their privacy policies and data protection measures, DeepSeek has not provided sufficient assurances that user interactions are safeguarded.
2. Potential Government Influence, Surveillance, and Propaganda
Given that DeepSeek is developed in China and the current competition between China and the US, there is legitimate concern that the model could serve as a tool for mass data collection and surveillance. Additionally, AI models have the potential to shape public opinion, and DeepSeek could be leveraged as a vehicle for state-sponsored propaganda. By subtly influencing responses to politically or socially sensitive topics, the model could be used to push narratives favorable to the Chinese government while suppressing dissenting viewpoints. This raises further concerns about the integrity of information disseminated through the platform and the potential for manipulation on a global scale. Similar accusations have been levied against TikTok, where critics argue that the Chinese government could exert influence over user data. If DeepSeek follows a similar trajectory, it may act as a Trojan horse for gathering global user data under the guise of an advanced AI model.
3. Unclear Legality of Training Data
Ethical AI development requires responsible data sourcing, but DeepSeek is not clear on whether its training data was legally obtained and whether it respects intellectual property rights. The AI community has already seen controversy surrounding unauthorized data scraping, leading to lawsuits and backlash against several LLM developers. Without transparency, users cannot be sure if DeepSeek’s responses are built on a foundation of legally acquired and ethically sourced data.
4. Uncertain Origin of the Underlying Model
Some reports suggest that DeepSeek may be leveraging OpenAI’s models, but there is ambiguity around whether this usage aligns with OpenAI’s terms of service. If DeepSeek has acquired or reverse-engineered OpenAI’s models without proper authorization, it raises both ethical and legal concerns. Unauthorized usage of proprietary AI technology could lead to regulatory action, further casting doubt on DeepSeek’s legitimacy as an independent innovation.
5. Lack of Adequate Guardrails
Unlike more established LLMs, DeepSeek lacks essential guardrails—safeguards implemented to prevent the generation of harmful, misleading, or unethical content. Guardrails are crucial in ensuring that AI models do not propagate misinformation, engage in biased outputs, or produce responses that could endanger or trigger users. Without these safeguards, DeepSeek presents a higher risk of misuse and unintended consequences.
6. Easily Circumvented Safeguards
Even where DeepSeek has implemented some form of moderation, security analysts have demonstrated that these safeguards are easily bypassed. Weak or ineffective barriers against malicious or unethical usage make the model particularly vulnerable to exploitation. This raises significant concerns for both individual users and organizations that might unknowingly expose themselves to liability or reputational damage by engaging with an unregulated AI tool.
Conclusion
AI innovation must be accompanied by transparency, ethical data practices, and robust security measures. At this stage, DeepSeek raises too many unanswered questions regarding privacy, legality, and trust. Until these concerns are adequately addressed, individuals and organizations should exercise caution before engaging with this platform. With a growing number of established AI models that prioritize responsible development, users have safer alternatives that do not carry the same level of risk.
Further Reading:
- We tried out DeepSeek. It worked well, until we asked it about Tiananmen Square and Taiwan
- DeepSeek: The Chinese AI app that has the world talking
- Chinese GenAI Startup DeepSeek Sparks Global Privacy Debate
- Italy Bans Deep Seek
- DeepSeek Failed Over Half of the Jailbreak Tests by Qualys TotalAI
- International regulators probe how DeepSeek is using data. Is the app safe to use?
- Be careful with DeepSeek, Australia says — so is it safe to use?