U.S. Warns Against Foreign AI: DeepSeek Lags in Tests
U.S. Commerce Secretary Howard Lutnick has raised concerns about the use of foreign AI models, particularly DeepSeek, which lagged behind American counterparts in recent tests. The models, developed by a Chinese company, showed potential biases and security risks.
The National Institute of Standards and Technology (NIST), under the U.S. Commerce Department, conducted a comprehensive study using its Center for AI Standards and Innovation (CAISI). The tests, which included benchmarks like SWE-bench Verified, Breakpoint, and MMLU-Pro, pitted DeepSeek models against leading U.S. systems such as OpenAI's GPT-5 and Anthropic's Opus 4.
DeepSeek models consistently underperformed, with OpenAI and Anthropic models outperforming them by 20 to 80%. Notably, DeepSeek models struggled in cybersecurity and software development tasks, raising concerns about their reliability for these critical areas. Additionally, the models showed a tendency to reproduce Chinese Communist Party narratives, sparking worries about political bias and censorship.
The study also highlighted potential security risks. Secretary Lutnick, in a post on X, thanked former President Donald Trump for his AI Action Plan, emphasizing the need to rely less on foreign AI to protect U.S. national security. DeepSeek R1, despite these concerns, continues to be adopted, potentially posing further risks.
The NIST study has revealed significant performance gaps and potential biases in DeepSeek models compared to their U.S. counterparts. With the models' continuous adoption and the raised concerns, U.S. officials are urging caution in relying on foreign AI, emphasizing the need for domestic innovation and security measures.
Read also:
- Germany's Coalition Explores Social Security Reforms Amid Record Spending
- chaos unveiled on Clowning Street: week 63's antics from 'Two-Tier Keir' and his chaotic Labour Circus
- Skechers Debuts First American Stores Focused on Athletic Footwear Performance
- Budget discrepancy jeopardizes highway projects' financial support
