00:00OpenAI recently reported that the Chinese government has used ChatGPT for surveillance
00:05and censorship. What risks do you see when an authoritarian regime uses AI tools like that?
00:11It's not just only because of the violation of human rights,
00:15but the potential to do the actual harm to people's lives, as we see in the Chinese
00:21targetting minority Uyghurs, and it's not just about within their countries, but they also
00:28utilize this kind of tool to targeting overseas Chinese communities, and that is a significant
00:36usage in the long arms and controls overseas. So that will be clear violations for the rule of
00:44laws and also the violation of human rights in this concept.
00:48The report also says that China-linked accounts use ChatGPT to draft phishing emails,
00:54targeting specifically Taiwan's semiconductor sectors. What do you think is the goal behind that?
00:59The nature for Chinese this kind of threat adversary actors, sometimes they are acting more like data
01:08brokers. Understanding the Taiwan's semiconductor sectors and how the Taiwan's response to the
01:15entire Taiwan's and U.S. trade negotiations, that's very important information too. So from China's
01:23perspective, all this data has the name tag and price on it, and that create the incentive, not just
01:31from national interest, but also economic incentive for this threat actor to specifically targeting
01:37standard conductor industry.
01:39For an AI company like OpenAI, besides exposing malign activities, what else do you think can
01:45be done to counter China's influence operations?
01:48I think the company is like ChatGPT because it's a large usage base right now. It's
01:56more has the kind of public obligation as the responsible actors to be transparent of the
02:06mind and adversarial usage. And what is the greatest challenge I will say for ChatGPT is that China
02:15obviously will try to measure the scope and scales of ChatGPT operations. And they will try to understand
02:25how ChatGPT responds to certain actions or certain questions that might consider as a harm to Chinese
02:33legitimacy. So that's a challenge to try to put that kind of restrictions, but also not limiting the
02:41language that they are legitimately used for other normal users.
Comments