In a year marked by seemingly weekly advances in AI capabilities, government authorities and lawmakers around the world have struggled to keep up. Starting next month, however, Chinese regulators will put in place new rules restricting one of AI’s most nerve-racking use cases: deepfakes.
On January 10, according to The South China Morning Post, China’s Cyberspace Administration will implement new rules that are intended to protect people from having their voice or image digitally impersonated without their consent. The regulators refer to platforms and services using the technology to edit a person’s voice or image as, “deep synthesis providers.”
Those deep synthesis technologies could include the use of deep learning algorithms and augmented reality to generate text, audio, images or video. We’ve already seen numerous instances over the years of these technologies used to impersonate high profile individuals, ranging from celebrities and tech executives to political figures.
Under the new guidelines, companies and technologists who use the technology must first contact and receive the consent from individuals before they edit their voice or image. The rules, officially called The Administrative Provisions on Deep Synthesis for Internet Information Services come in response to governmental concerns that advances in AI tech could be used by bad actors to run scams or defame people by impersonating their identity. In presenting the guidelines, the regulators also acknowledge areas where these technologies could prove useful. Rather than impose a wholesale ban, the regulator says it would actually promote the tech’s legal use and, “provide powerful legal protection to ensure and facilitate,” its development.
But, like many of China’s proposed tech policies, political considerations are inseparable. According to the South China Morning Post, news stories reposted using the technology must come from a government approved list of news outlets. Similarly, the rules require all so-called deep synthesis providers adhere to local laws and maintain “correct political direction and correct public opinion orientation.” Correct here, of course, is determined unilaterally by the state.
Though certain US states like New Jersey and Illinois have introduced local privacy legislation that addresses deepfakes, the lack of any meaningful federal privacy laws limits regulators’ abilities to address the tech on a national level. In the private sector, major US platforms like Facebook and Twitter have created new systems meant to detect and flag deepfakes, though they are constantly trying to stay one step ahead of bad actors continually looking for ways to evade those filters.
If China’s new rules are successful, it could lay down a policy framework other nations could build upon and adapt. It wouldn’t be the first time China’s led the pack on strict tech reform. Last year, China introduced sweeping new data privacy laws that radically limited the ways private companies could collect an individual’s personal identity. Those rules were built off of Europe’s General Data Protection Regulation and turned the volume up to 11. Like the new deep syntheses or deepfake regulation, last year’s privacy rules specifically required companies to obtain consent before collecting personal data. For sensitive data like a person’s fingerprint or financial details, providers are required to obtain consent yet again.
That all sounds great, but China’s privacy laws have one glaring loophole tucked within it. Though the law protects people from private companies feeding off their data, it does almost nothing to prevent those same harms being carried out by the government. Similarly, with deepfakes, it’s unclear how the newly proposed regulations would, for instance, prohibit a state-run agency from doctoring or manipulating certain text or audio to influence the narrative around controversial or sensitive political events.