AI Security
DeepSeek's Hidden Risks: Should You Be Concerned?

Popularity vs. Privacy DeepSeek R1's Rapid Rise and the Red Flags It Raises

DeepSeek R1, a free and open-source AI assistant, has taken the tech world by storm, becoming the most downloaded free app on the Apple App Store. Its affordability and efficacy, as perceived by many users, have fueled its rapid adoption.

However, this swift rise has prompted security experts and AI professionals to take a closer look. Their findings reveal several significant concerns that potential users should consider before integrating DeepSeek into their workflows. Is the convenience worth the risk?

Data Control 1. DeepSeek's Data Retention: A Privacy Nightmare?

Heather Murray, an AI consultant for major corporations and the U.K. government, expressed serious concerns about DeepSeek's data retention policies. According to Murray, DeepSeek retains user data indefinitely, even after users leave the app.

This data, transmitted and stored on servers in China, falls under Chinese law, potentially exposing users to different legal standards than those in the U.S., U.K., or Europe. Running DeepSeek locally on a personal desktop (offline) is a safer alternative, but many users may prefer the convenience of the cloud.

Privacy Policy 2. Keystroke Tracking: DeepSeek's Red Flags

Tara Tamiko Thompson, an AI educator and advisor, highlights a critical concern within DeepSeek's privacy policy: keystroke tracking. This practice allows DeepSeek to monitor and record everything you type while using the application.

Thompson emphasizes the potential dangers of keystroke tracking, which could lead to biometric hacking, behavioral profiling, social engineering, and other cyber threats. The assumption that app store presence guarantees regulatory compliance can be dangerous.

“

It keeps your data as long as it wants to, and even after users leave the app, it doesn’t delete their data. It’s going to hang on to that. That is a massive worry.

Heather Murray, AI Consultant

Censorship Concerns 3. DeepSeek's Censorship: Biased Information Retrieval?

Chris Duffy, founder of Ignite AI Solutions, raises concerns about censorship within DeepSeek, particularly due to its Chinese origin. AI models trained in China must comply with strict regulations that prevent discussion of politically sensitive topics.

Duffy demonstrated DeepSeek's censorship by querying the system about Chinese government tactics to control online content. The model initially refused an output but provided an answer when the query was submitted as an image, only to censor itself seconds later. This raises concerns about biased information retrieval for global businesses and researchers.

Energy Use 4. DeepSeek's Hidden Costs: Not Always Cheaper for Enterprises

Despite being touted as a more efficient AI model, testing from Arthur D. Little suggests that DeepSeek's chain-of-thought reasoning leads to significantly longer outputs, potentially increasing total energy consumption. While per-token efficiency may be better, the increased output can negate these gains.

Michael Papadopoulos, an ADL partner, advises organizations to carefully evaluate DeepSeek's open-source models alongside other options, with clear guardrails for potential bias and security. He also suggests avoiding DeepSeek's official hosted services due to unresolved privacy, security, and regulatory risks.

“

As soon as I saw it reference — plain as day — that they monitor keystrokes, I was out. I’m shocked others don’t feel the same way.

Tara Tamiko Thompson, AI Educator and Advisor

Conclusion Think Twice Before Diving Deep Into DeepSeek

DeepSeek's surging popularity is undeniable, but the red flags raised by experts – from questionable data practices to keystroke tracking – suggest users should proceed with caution. Evaluate your own risk tolerance and data security needs before integrating this AI assistant into your workflows.

Company representatives didn't respond to requests for comment on these concerns, leaving users to weigh the risks and benefits based on available information. Is DeepSeek's functionality worth the potential privacy trade-offs?