According to Bloomberg's report on October 14th, Sarah Flare, the chief financial officer of OpenAI, has caused trouble. Companies that choose to use low-cost Chinese artificial intelligence models are taking on cybersecurity risks.
Her statement resonated with the anxieties of Anthropic's competitors.
Fleer said in an interview during the International Artificial Intelligence Summit held in Dublin, Ireland on Wednesday: "To those who might be adopting Chinese models, I want to say that you should also consider issues of security and safety. From a cybersecurity perspective, we believe it's better for you to use Western models."
As open-weight artificial intelligence models continue to make progress, especially those from China, OpenAI and Anthropic are under pressure. These models allow users to download the model parameters and run them on their own hardware devices. Compared to the top models offered by American companies, these models are usually cheaper, and their performance in certain tasks is almost equivalent.
Reports indicate that an increasing number of developers are turning to these models. On the model hosting platform Hugging Face, models developed in China account for the largest share of downloads of generative artificial intelligence models.
OpenAI and Anthropic mainly provide closed-source models that cannot be downloaded or modified. Both companies have been trying to present open-source models as a riskier option.
"During an interview at the same event earlier today, Anthropic's International Business Head, Ji Yao-mu Pulan, stated: 'We must be extremely cautious in dealing with open-source models on safety issues. I cannot speak for open-source models, but from a definition standpoint, they carry higher risks.'"
Feller said during her speech at the summit that the progress made by Chinese artificial intelligence models means that American companies should not overly slow down the release of these models. She added that OpenAI has delayed the release of some models to ensure they meet the company's internal security standards.
Fleer said, “I do believe that we need to control the pace of development, but we cannot slow down to a point where there is an imbalance in who can access these tools. From a cybersecurity perspective, these tools could lead to significant issues related to critical tasks worldwide.”
This summer, OpenAI's models broke through the testing environment limitations and invaded the Hugging Face platform, causing new concerns about artificial intelligence cyberattacks. After further investigation, the company revealed other security incidents, including government websites being hacked. Finally, Chinese open-source AI models participated in security analysis and defense efforts.
The Chinese side has repeatedly emphasized that China always attaches equal importance to development and security. It places great emphasis on various inherent and derivative risks posed by artificial intelligence, and is committed to establishing a solid safety baseline. The Chinese side continues to improve laws and regulations, policies, application standards, and ethical rules, in order to prevent the abuse or misuse of artificial intelligence, and to ensure its safety, reliability, and controllability.
