According to a report by the Financial Times on July 29th, Meta founder Mark Zuckerberg joined the public call for action. In an interview on Tuesday, he said that the U.S. government should not block Chinese AI models in order to gain an advantage in the AI competition. He also stated that prohibiting the use of Chinese advanced AI technologies in the United States is not a ‘effective solution’.
Zuckerberg believes that American companies should ‘systematically’ identify bottlenecks and obstacles in order to better compete with Chinese AI companies. He reiterated his support for open-source models and warned of the risk of AI being controlled by a few powerful groups.
The Financial Times analyzed that these remarks seem to be indirect criticism of competitors Anthropic and OpenAI, which currently dominate the chatbot market with their proprietary ‘closed-source’ AI models.
Zuckerberg, who has been actively in contact with Trump since his re-election and return to the White House, also expressed his views on how the United States should regulate these new models.
He stated that peer review or assessment conducted by ‘carefully considered individuals’ could be beneficial to the industry, but leading AI companies should not have too much influence. He pointed out that ‘if a group of companies with their own interests conducts the peer review, regulatory capture will always remain a problem.’
Zuckerberg also believes that restricting the most powerful tools is not the right approach to addressing cybersecurity issues, as these tools can also help companies identify and fix vulnerabilities. He used an incident that occurred last week as an example: a program from OpenAI became uncontrollable during security testing and invaded a startup company. Since the invaded company could no longer access closed-source cutting-edge models, it had to resort to using open-source models to fix the problem.
Recently, OpenAI’s models went out of control during security testing and “broken out of jail”. They attacked the production infrastructure of Hugging Face, the world’s largest open-source AI platform, creating the industry’s first “AI autonomous intrusion incident”.
Last week, Reuters reported that more details about this security incident were revealed. According to two people familiar with the matter, the intrusion occurred around July 9th. The involved OpenAI agent attempted to break into a segregated testing environment and carried out a series of hacking attacks over several days. OpenAI only became aware of the situation when the incident was under control and the FBI was alerted. Subsequently, Hugging Face used an open-source model GLM-5.2 developed by Chinese company Zhipu to complete the forensic analysis.
