Open Source really works. They open sourced the Diia app recently, their super app. It’s really flipped the default because many people in national security or defense and so on, they see open source and they think the adversary can send pull request, which is scary. But when coupled with ...

Yeah, Microsoft, Google, and Amazon.

They don’t even deny it. They just say it’s an accident. I think with the gray zone operations, that is going to be the norm. We have to assume breach for our physical infrastructures that are location-transparent to the entire society.

But actually, we found a lot of attacks are those very attrition-type-like denial of service attacks on availability, like the lines between Mazu and Taiwan were physically cut by fishing cargo vessels flying the PRC flag.

If we are locally resilient and we convince the three public cloud providers to set up data centers in Taiwan, which they all are committed to for this year, and for them to work with OneWeb and SES, and other resilient networks, then we have a pretty really good backup ...

Frankly, most of the private sector, especially the three large cloud vendors, already do some part of it as part of their work. One of the lessons we learned from Ukraine is that if the cloud providers is willing to set up their stuff in a locally resilient way, including ...

Yeah, we just met people from Paraguay last night. With their technology minister.

But also we started later. People who started earlier have to pay more for this kind of thing. But now, like FIDO implementations, interoperable protocols around zero knowledge verification, what the EU calls digital identity wallets, and so on. These are now mature technology, thanks to blockchain. Now, when we ...

Anyway, so by insisting on interoperable protocols, transparently procured, we are making it a part of the procurement process. We are prototyping it with the TSMC and semiconductor manufacturing sector, and this year spreading to the rest of the OT sectors, including energy and water and electricity and so on. ...

But the idea is that when I sign official documents, there’s no password anymore. Biometric is verified by this on device final implementation, the device integrity verified by crowdstrike, the behavioral pattern analyzed by a cloud layer, and we never go to the same vendor for two adjacent layers, not ...

Working with TSMC and Friends, we published an international standard, SEMI E187, applied zero trust principle in manufacturing, especially around semiconductives. That actually rolled out even before we fully secure our government’s Class A, meaning holding the national person data, competent authorities, all the ministries, 47-ish, is already rolling out ...

The TSMC did learn from the cyber attack attempts on it that it is not just TSMC proper, but if any weak point in the supply chain gets exploited, the entire supply chain is at risk. They adopted a what’s called assume breach mentality, assuming that the adversary has already ...

They are a high-value target. Since we’re going to co-edit the transcript, I’m going to say that it’s even more valuable than government’s data for cyber attacks.

Yeah. We have a very special manufacturing sector located within a critical infrastructure domain called Science Park, which is a very Taiwanese thing, and that is the TSMC and its supply chain.

Volt Typhoon, all sorts of typhoons.

So with this rapid coordination and rapid legislation, I think we’re still in a phase where it’s manageable because we took action before it got to community spread, so to speak. But if it’s already widely community spread, then maybe some lockdown or takedown will be filled as justified. But ...

So think of it like a radar scanning of what people across all parties feel as the most immediate mark and what is considered proportionate for us to assign a liability scheme. And once that is collected, we send it to the department and say, there’s this is special anti-scam ...

We sent these to hundreds of thousands of people. We use a sortition method, chose a representative sample of 450 people, and asked them to go online this Saturday, on a series of deliberations about the red lines of the platforms and how much state control is warranted for which ...

This is also something that is very helpful. Instead of asking people to look at content or behavior of communication, just have short codes, easily memorable, and say everything else that purports to be official is a scam.

This is also why our investment scam law is the way it is. There’s no fine on Facebook, but if somebody get conned for 1 million after Facebook knowingly caused that advertisement of defake, Facebook is liable for that 1 million. We’ve not fined Facebook yet, but they actually complied ...

That’s a great question. I think it’s manageable if and only if we fully ask the largest platforms to internalize the negative externalities and harm it caused. Without this liability arrangement, there’s no amount of watermark that can work alone. That doesn’t work. The only way it works is just ...

I think it’s also very helpful in inoculating people against this shapeshifting doppelganger attacks. These two are my main suggestions for US or any other jurisdiction looking to defend an election. That and also paper-ballot only and participatory monitoring of the counting process, but you already know that.

So this is not just a scam factor, but also a precision persuasion factor. We’ve worked, now that we’re about to become a competent authority for our online advertising platforms, we’re now working closely with platforms such as Facebook and Google to make sure that we switch to a new ...

Starting this year, we’re going to roll it out to all advertisements, not just political ones, because we’ve seen that convincing AI chat bots imitating celebrities, they can start with a very wide-ranging advertisement and then lure people into one-on-one conversation with certain celebrity in real time. And armed with ...

The other thing is requiring KYC for political advertisements. We’ve had an agreement in 2019 with Facebook’s civic integrity team, that said on political advertisements need to carry KYC and foreign funding sourced are prohibited, just like campaign funding.

Instead of an anti-vax faction, which we don’t have in Taiwan, we have people who vehemently believe that Medigen is just water, and only BNT has value or Moderna or AstraZeneca. I got four different brands of vaccine myself just to make a point. Anyway, so that’s pre-bunking.

I’ve also filmed a deep fake of myself to pre-bunk such manipulations. There’s many tactics that you use here. We pioneered that during the pandemic where we turned this vax/anti-vax information manipulation — which you may be familiar with — into a friendly competition between the four brands of vaccines.

First is pre-bunking. That is to say, we make sure that the information manipulations, even before they go viral, we make sure that the parties are in the process in the discovery of such manipulations. It’s not a partisan thing, but rather it’s a civil society thing. When people learn ...

The main metric we measure is affective polarization. That is to say, whether people love each other or hate each other when they belong to different political parties. We are happy to report that polarization, when measured this way, decreased after the election, which is quite unusual, if I say ...

AI is very much part of this landscape. In our January election, we saw not just the traditional denial of service attacks, which increased by 33 times compared to the previous year in the same quarter. But also information manipulation for the first time, convincing voice claims, convincing social media ...

Great.

Awesome questions.

This capability that I just mentioned.

But the Taiwanese people are very sensitive to anything like this. Now, all this is in the cabinet and soon will be for the legislature to deliberate.

But I think there’s a limit of doing things this way. We had to complement the RPZ capability with radical transparency, with publishing as open data, where the request came from, with the appeal process. If you connect to the blocked website, you can see this email that you can ...

They want to make sure that we can, for example, block entire domain names, block entire apps, and so on.And to these people, there’s part of it that we are already doing, the DNS RPZ mechanism, which does allow us to block entire Internet domains if that entire domain is ...

And so I’m reasonably sure that the measures that we have, because it’s partly drawn from their consensus, meets this microcosm of people. On the other hand, I also fully understand there are people of stronger views, like, they may not be statistically representative of Taiwanese, but they see dangers ...

We work with Stanford on that. And I think around 500 actually did show up that day. 450 stayed for the entire session, which is amazing in terms of engagement rates anywhere in the world.

This is the deliberative, the deliberative polling.

Well, according to the… I think it’s a microcosm of 450 people… we sent to 170,000 random SMS accounts. And of the people who wanted to attend, I think 600 people we chose based on this microcosm statistic representativeness, and more than 580 said that they wanted to participate.

Yeah.

Because the threat landscape has changed dramatically.

Very much so.

Yeah, of course.

Yes. And for 10 years, the top target for disinformation attacks, according to V-Dem.

Yes.

And now, with the online fraud also defrauding people, as soon as new harm is surfaced or speculated, we will move very quickly to amend that part of the law, instead of waiting on a comprehensive AI act to anticipate all harms, which is probably not going to be possible.

So, these are the main things that worries me. But for OpenAI, Anthropic and so on, we’re part of this Alignment Assemblies project, and so, including info integrity, Alignment Assemblies, which was run last Saturday, is just to draw boundaries, so that people can surface new harms, or new potential ...

Which is why I think this kind of amplification, we might see it more and more. And so, whether a platform is bona fide adhering to those transparency and accountability laws, or whether they sign on it, but like a reserved capability to do like this kind of strike, will ...

So, one video can look at like 10 different captions, or 10,000 or 10 million different captions, depending on how many people are viewing that video, and each generated caption would fit that person’s reward model to maximize effect on any axis, really. So, this is the kind of interesting ...