China Mandates Security Reviews for AI Services Like ChatGPT

(Bloomberg) -- China plans to require a security review of generative AI services before they’re allowed to operate, casting uncertainty over ChatGPT-like bots unveiled by the country’s largest tech companies including Baidu Inc.

Most Read from Bloomberg

• Katie Cotton, Guardian of the Apple Brand for 18 Years, Dies

• Twitter Company ‘No Longer Exists,’ Is Now Part of Musk’s X

• Elizabeth Holmes Will Have to Wait Out Her Appeal in Prison

• Inflation Data Will Shatter the Stock Market Calm, Goldman Partner Warns

• Tumbling Money Supply Alarms Economists Who Foresaw Inflation

Providers of services must ensure content is accurate and respects intellectual property, and neither discriminates nor endangers security, the Cyberspace Administration of China said in draft guidelines published for public feedback. AI operators must also clearly label AI-generated content, the country’s internet overseer said in a statement posted on its website.

The CAC’s requirements add to Beijing’s growing attempts to regulate the explosive growth of generative AI since OpenAI’s ChatGPT fired up the industry in November. Companies from Alibaba Group Holding Ltd. to SenseTime Group Inc. and Baidu all aim to build the definitive next-generation AI platform for the world’s largest internet market. That mirrors a growing wave of development abroad with Alphabet Inc.’s Google and Microsoft Corp. among the many tech companies exploring generative AI, which can create original content from poetry to art just with simple user prompts.

China’s made no secret of its wish to elevate AI at a time the country is locked in a conflict with the US over technology from chips to EVs. But it remains uncertain how the government intends to both galvanize and police the emergent field. Tuesday’s guidelines essentially port existing regulations on data and content — from the protection of personal information to censorship of messages the Party deems undesirable — to the burgeoning field of AI.

China will probably bar foreign AI services, like those from OpenAI or Google, as it did with American search and social media offerings. Beijing insists on tight control over online content and discussion. But Beijing is expected for now to avoid tightening the leash on its local firms too much, for fear of squelching a nascent arena that needs room for innovation, said Angela Zhang, Associate Professor of Law at the University of Hong Kong.

“The CAC’s quick reactions to this new technology clearly demonstrates its regulatory ambition in this sphere,” Zhang said. “These developments will likely have spillover effects on Chinese AI regulation in the future. Thus far, however, I see the Chinese regulators being quite cautious with its regulatory approach in order to give more room for the development of generative AI in the country.”

Alibaba shares gave up much of their initial gains on Tuesday after the CAC announcement, while SenseTime was largely unchanged. The larger e-commerce company on Tuesday described how it planned to build generative AI into its Slack-like work app and Amazon Echo-like smart speakers, before expanding that portfolio to its other services. A day before, SenseTime demonstrated the large AI model SenseNova and a user-facing chatbot called SenseChat.

That followed Baidu Inc.’s Ernie bot, which was released for selective testing about a month ago. The company — considered the current domestic leader — slid as much as 7% in Hong Kong. Chinese carriers such as China Mobile Ltd. and smaller AI-linked stocks from 360 Security Technology Inc. to AI chipmaker Cambricon Technologies Corp. were also down sharply.

“There’s real potential there to affect how the models are trained and that stands out to me as really quite important here,” said Tom Nunlist, a senior analyst at Trivium China. The rules will likely impact the way that AI models in China will be trained going forward, although much will depend on how regulators interpret broad provisions in their notice, he added.

Read more: Alibaba Enters ChatGPT Fray With AI Speaker, Slack-Like App

Beijing plans to introduce rules to govern the use of artificial intelligence across a swath of industries. In February, officials from the Ministry of Science and Technology said the government will push for the safe and controllable application of AI services, which it considers a strategic industry.

In addition, the powerful regulator stressed that AI services must be transparent about the data and algorithms used in training their large-scale models, reinforcing Beijing’s focus on maintaining control over sensitive and valuable information.

“Service providers should provide certain information on data used in AI training, including origin, size and type of data,” the CAC said in its statement. “It also requires AI platforms to share basic algorithms and other technologies used.”

What Bloomberg Intelligence Says

Beijing’s heightened regulatory scrutiny of ChatGPT-style services in the country, based on new guidelines issued by the Cyberspace Administration of China (CAC) on April 11, looks unlikely to deter expansion plans by Alibaba, Baidu, SenseTime and NetEase in this field, we believe. CAC’s proposed rules, including public censure for first-time violations, and suspension of operations and fines for repeat offenders, are in line with general directives already in place to govern local tech firms.

- Catherine Lim and Trini Tan, analysts

Click here for the research.

Taken together, the rules may be intended to ensure ChatGPT-like services are subservient to the Communist Party’s rigid ban on controversial or undesirable content online. But it could also boost companies like Baidu and SenseTime by providing clearer ground rules for future services. On Tuesday, the CAC said content considered deviant from society will be prohibited, without elaborating.

It’s unclear how the new regulations will affect AI development going forward. Just about every major tech firm including Tencent Holdings Ltd. is exploring ways to harness its potential.

There are concerns however over whether Chinese companies can secure reliable access to the high-end chips and technology needed to develop large-scale AI models over the longer term. SenseTime, for example, is operating under US sanctions that inhibit its access to capital as well as crucial American components, and the Biden administration last year imposed restrictions on the sale of AI accelerator chips to Chinese customers — a critical component in the development of any large-scale generative model.

--With assistance from Sarah Zheng.

(Updates with analyst’s comment and share action from the fifth paragraph)

©2023 Bloomberg L.P.