top of page
13 Logo With Text Colored.png

advertisement

GCFB_40th_logo_bright.png

The vital link between
food & hunger

AI Theft or Common Practice? OpenAI Investigates DeepSeek; Reports Show Alarming Tech Security Flaws

Writer's picture: Toni MathewsToni Mathews

CLEVELAND 13 (WCTU) — OpenAI has accused Chinese artificial intelligence startup DeepSeek of using its proprietary models to develop a competing open-source AI system, raising significant concerns over potential intellectual property theft and national security risks. The allegations, which OpenAI has been investigating for months, center around DeepSeek’s suspected use of "distillation", a process that allows developers to train smaller models using outputs from larger, more powerful ones.


The San Francisco-based AI company, known for its ChatGPT product, stated that while distillation is a common industry practice, DeepSeek’s alleged use of it to create a direct competitor would constitute a violation of OpenAI’s terms of service. These terms explicitly prohibit users from “copying” any of OpenAI’s services or using its output to develop competing models.


Mounting Evidence and Investigations

OpenAI has yet to release definitive proof but claims to have seen substantial evidence that DeepSeek engaged in distillation to replicate its models. "The issue is when you take this output and use it to create your own model for your own purposes," a person close to OpenAI told the Financial Times.


Microsoft, OpenAI’s largest partner, reportedly blocked accounts believed to be associated with DeepSeek last year after suspecting that the startup had violated OpenAI’s terms of service. These investigations were first reported by Bloomberg, but OpenAI has remained relatively tight-lipped about the extent of the potential data breach or the exact methods DeepSeek used to train its AI.


The claims have surfaced at a time when U.S.-China tensions over AI technology are at an all-time high. DeepSeek’s rise has shocked Silicon Valley, as its models—trained with significantly lower costs than those of OpenAI or Google—have already demonstrated performance on par with leading U.S. models.


The allegations have also prompted speculation on Wall Street. On Monday, Nvidia shares dropped 17%, wiping $589 billion off its market value, amid fears that cheaper AI models could reduce demand for expensive AI hardware. However, tech stocks rebounded on Tuesday, as analysts downplayed the long-term impact of DeepSeek’s advancements.


DeepSeek’s Meteoric Rise and Its Response to Allegations

DeepSeek has not publicly responded to OpenAI’s accusations. The timing of these allegations coincides with DeepSeek’s recent AI breakthroughs, including the release of its R1 reasoning model, which has outperformed expectations and positioned the startup as a serious contender in the AI race.


Industry experts remain divided over whether DeepSeek could have truly trained its large language models (LLMs) with the reported budget of just $5.6 million using 2,048 Nvidia H800 GPUs. The cost is a fraction of what OpenAI and Google spend to train similarly sized models, raising questions about the true origins of DeepSeek’s capabilities.


Some experts argue that the model’s responses suggest it was trained on OpenAI's GPT-4 output, which would directly violate OpenAI’s terms. "It is very common for startups and academics to use outputs from human-aligned commercial LLMs, like ChatGPT, to train another model," said Ritwik Gupta, a PhD candidate in AI at the University of California, Berkeley.


Other analysts, however, suggest that DeepSeek’s achievements are "evolutionary rather than revolutionary." Vivek Arya, an analyst at BofA Securities, pointed out that DeepSeek's models were likely built upon Meta’s open-source Llama models and that its novel approaches were not necessarily groundbreaking.


Security Risks and Data Privacy Concerns

Beyond the potential IP violations, DeepSeek has faced growing scrutiny over its handling of user data. Recent reports revealed that a publicly accessible DeepSeek database had been exposed, leaking highly sensitive information, including chat histories, secret keys, and backend operational details.


Cybersecurity firm Wiz Research found that DeepSeek’s database, hosted on unprotected servers, allowed full control over database operations, making it possible to access over a million lines of log data. The exposure included API secrets, timestamps, and user queries. While DeepSeek quickly secured the vulnerability after being alerted, the breach raises concerns over the security of AI models developed without stringent data protection measures.


"AI companies are moving at breakneck speeds, but they are not prioritizing security," said a Wiz Research spokesperson. "This case is a stark reminder that even the most advanced AI models are vulnerable to breaches."


DeepSeek’s privacy policy also states that user data, including conversations and generated responses, is stored on servers in China, triggering concerns over potential government access. Adrianus Warmenhoven, a cybersecurity expert at NordVPN, warned that China’s data laws allow the government to compel companies to share user information.


"The concern is not just data privacy, but how that data could be used," Warmenhoven said. "We are in an era where AI models are becoming central to information access, and if a government has unrestricted access to that data, the implications are vast."


Parallels to the TikTok Ban and National Security Risks

The DeepSeek controversy has reignited discussions about the U.S. government’s scrutiny of Chinese technology firms. Similar concerns were at the heart of the ongoing efforts to ban TikTok, the Chinese-owned social media giant, over fears that the Chinese Communist Party could access American user data.


TikTok has repeatedly denied these allegations, but the U.S. government has not been convinced. Lawmakers have cited China’s cybersecurity laws, which require companies to cooperate with national intelligence efforts, as justification for potential bans or forced divestitures of the company.


DeepSeek’s alleged misuse of OpenAI’s models adds another layer to these concerns. Unlike TikTok, which deals with user-generated content, DeepSeek is developing AI models capable of information processing and decision-making at a scale previously unseen. If China-based companies can train their models using U.S. technology, it could accelerate China’s AI development and weaken the competitive advantage of U.S. firms.


The White House has already signaled that it is closely watching AI developments in China. "We know China-based companies, and others, are constantly trying to distill the models of leading U.S. AI companies," OpenAI said in a statement. "We believe it is critically important that we work closely with the U.S. government to protect these capabilities from adversaries and competitors."


The Future of AI Regulation and Intellectual Property Protection

As OpenAI deals with allegations of its own copyright violations—including lawsuits from The New York Times and various authors—the company now finds itself on the other side of an intellectual property battle.


The case highlights the broader challenge of regulating AI technology, particularly in international markets where legal frameworks around intellectual property protection remain inconsistent.


"The AI arms race is happening now, and the rules are not keeping up," said John Scott-Railton, a senior researcher at the University of Toronto’s Citizen Lab. "What we are seeing with OpenAI and DeepSeek is a preview of future battles between major AI players."


For now, the U.S. government has not taken direct action against DeepSeek, but policymakers are increasingly calling for stricter measures to prevent foreign entities from exploiting U.S. technology.


As AI becomes an essential part of daily life, the fight over who controls the most powerful models will only intensify. Whether DeepSeek’s alleged actions prompt stronger legal protections for AI companies remains to be seen, but one thing is clear—this case is far from over.

Commentaires


join C13 banner.jpg

advertisement

bottom of page