OpenAI Alleges DeepSeek Violated AI Safeguards, Bypassed Restrictions

TLDR

• OpenAI accused Chinese AI startup DeepSeek of bypassing access restrictions to replicate US-developed AI models.
• DeepSeek employees reportedly used obfuscated third-party routers to access OpenAI’s models and distill outputs for their own use.
• OpenAI highlighted concerns about the political bias and censorship found in DeepSeek’s AI models, including issues with sensitive topics.
• The memo pointed out that DeepSeek’s actions posed a risk to AI security and the integrity of US technological developments.
• OpenAI warned that unauthorized distillation without proper governance could lead to AI systems lacking necessary consumer protections.

In a recent memo to the US House Select Committee on Strategic Competition between the US and China, OpenAI accused the Chinese artificial intelligence startup DeepSeek of bypassing access restrictions on American AI systems. OpenAI claims DeepSeek used unauthorized methods to replicate and distill outputs from US-developed AI models, including those of OpenAI itself. The memo sheds light on ongoing tensions between the two nations in the race for AI dominance.

OpenAI Claims DeepSeek Is ‘Free-Riding’ on US AI Developments

OpenAI stated that DeepSeek, a Chinese competitor based in Hangzhou, has been systematically using US-made AI technology without proper authorization. According to the memo, the Chinese startup has been engaging in “distillation,” a method of training smaller models using outputs from advanced AI systems. OpenAI said that DeepSeek employees developed techniques to bypass safeguards and access its models through third-party routers and other obfuscated methods.

The company described the practice as unethical, with DeepSeek exploiting US innovations for its own gains.

Allegations of Censorship and Political Bias in DeepSeek’s AI Models

Beyond technical concerns, OpenAI also pointed to DeepSeek’s content governance practices. The memo accused the Chinese startup of imposing political bias within its AI systems. OpenAI found that the DeepSeek language model exhibited a strong pro-Chinese Communist Party (CCP) bias, often censoring sensitive topics such as Tiananmen Square or Taiwan independence.

When asked about topics critical of the CCP, such as the Falun Gong, DeepSeek’s model refused to provide answers, often deleting responses. OpenAI noted that such censorship practices raised serious questions about the potential impact of AI systems without sufficient safeguards. “Distilling US AI technology without proper governance could lead to systems that prioritize censorship over open dialogue,” OpenAI warned.

US and China Compete for AI Leadership Amid Resource Constraints

In its memo, OpenAI also highlighted the importance of computational resources, specifically the AI chips needed to power advanced systems. OpenAI stressed that the US holds an advantage in AI due to its access to powerful chips and the ability to generate electricity on a large scale. However, the memo raised concerns that China’s DeepSeek could eventually obtain access to these resources, which could shift the balance in AI competition.

The US has already begun investigating DeepSeek’s use of US-made hardware, with reports indicating that the Chinese startup may soon be able to purchase Nvidia’s H200 AI chips. These chips are critical for running complex AI models, and DeepSeek’s acquisition could further escalate the technological race between the two countries.

The post OpenAI Alleges DeepSeek Violated AI Safeguards, Bypassed Restrictions appeared first on CoinCentral.