如果還沒有商轉的話，我覺得可以先當一個 PoC，等到你們以後真的有 FHE，我們再把真正國家機密、軍事機密放上來，還是要分層，大概是這樣。

我記得你們還有另外一個完全對加密資料去進行運算的全同態加密，不過看各位的表情應該是還沒有商轉，還在研究室裡面。我這次去以色列有一個新創，他們主張的是就像 AI 有專門的 TPU，他們主張全同態加密也要有 3PU。因為我現在對著你的 CPU 加密送出去，我還是要相信你跟 CPU 的製造商沒有勾結，但是如果是全同態加密（FHE），我可以跟我的 auditor 說怎麼樣都不能還原。

有點類似 AMAZON 的立場，Outpost 能賣就賣，不能賣就不要賣，兩個都有最好。

但是你們有偏好嗎？維然講的是實體混合雲，我在講的比較是用密碼學的方式，是用模擬出一個租借，你們覺得哪種比較好？

這個需求之前微軟也有來談過，他們提議的解決方案就是我們自己塞自己的 Key 進去，他們保證不會在記憶體解密的時後偷看，所以概念上是我們送去運算的部分是直接加密到晶片，他就主張落地，這就跟放在自己家一樣，我不知道你們有沒有現在的 offer，你們對這樣的想法是什麼？

所以沒有這個需求？

包含內政部的公有雲服務，他們要直接這樣用嗎？或者他們去用 GCP 了？

這個是誰的需求，我們是要滿足誰的需求？

So, we’re talking about this hybrid Ccoud configuration where each government agency may have some locally-kept data in their local compute, but for certain compute that is beyond the local compute capability, they may, through dedicated lines, send to the Google data centers for additional computing. But the source of ...

有些運算放地端，不是所有的東西一定要到 Google 的機房算？

這個有點像是我們以前在做公部門混合雲的那種東西，對不對？

Sure. Mandarin is fine. I can interpret for you. You can do the whisper…

So, what is this PoC that you’re talking about, if not our Google Workspace?

And the second is that the physical apparatus, the data center, need to be administered by people with our nationality. This is easily satisfied by Google, more so than pretty much any other data center company. So, this, I’m not worried. So, when you say that you have a PoC, ...

In Taiwan, actually, our Personal Data Protection Act do not forbid extra jurisdictional processing. So, when we say local resilience, we don’t mean that you cannot send the private data you collected here to other places to compute. What we do mean is that when, not if, all the subsea ...

Okay, great. So, just to reiterate, here we use the word local resilience instead of data localization, because data localization already means something else. We understand that many jurisdictions use data localization to mean data residency, which means that according to their privacy law, anything that’s collected within the jurisdiction ...

Great. Thank you. It’s a wonderful agenda. Any from your team that would like to add?

That is fine.

Yeah, when is it coming to our Workspace account?

What to not do.

No, that’s fine.

Okay. Great. So, it’s…

Okay.

In the Department of Digital Service.

Yes. So, and welcome. Perhaps a very quick round of self-introduction from our side?

Thank you.

Thank you.

Yes. Of course.

Look, Ku Chuan — because she’s really looking at me now — I can get you the actual numbers after this interview.

The budgets of MODA and NCC combined…

Yeah, it’s around… It’s like a MEO receiver. You can think of it as like your home broadband. So, if it’s just you and your family using it, that’s okay for video conferencing, maybe not for esports gaming because of the high latency. But if you start sharing with your ...

Crazy slow. Because as I mentioned, they did not have a local cloud center. If they had a local data center and reroute the local to national domestic communication through that center, then it would not be slow.

The microwave was supers slow.

That depends on the repairing vessel’s schedule. Usually, it takes a couple of months or so.

Currently, microwave is barely enough. But by the end of the year, microwave will be plenty.

No, it’s not very deep.

Well, the details, I think the NCC has more details than I do, but what I know is that they’re both flying PRC flags and they both say it’s an accident.

And they said it’s accident.

Yes.

But we are investing in another additional cable between Taiwan and Matsu just because plurality. So, I think the main lessons we learned is psychological, in how to manage the expectation on reduced bandwidth, how to prioritize the bandwidth use, which uses are OK to have a slightly higher latency, ...

Yeah, so the microwave transmitters by the end of this year will have this capacity to satisfy Matsu’s domestic needs. So, we learned, of course, that any place that is well known can be disrupted, so the investment on microwave is necessary. We also learned that SES Global’s MEO actually ...

Yeah, so as I mentioned, there was a spike last August. And last year and the year before that, we saw much more coordinated attacks and also higher degree of attempted attacks and so on compared with 2020. So, 2020 and every year before it was like one background level. ...

It was getting higher significantly in the past couple of years. So, like this. This year, we’ve not had a complete pail yet, so maybe we talk by the end of the year.

But that’s quite similar to cyber-attacks as well because for many of those cyber-attacks, it’s not clearly attributable. So, when we say they’re of a foreign origin, we’re not being… like this is not a euphemism. We literally only know it’s of a foreign origin.

You’re like, yes, maybe we sign a peace accord with earthquake? No, this is not going to happen with the “地牛”, I’m sure…

And Ukraine.

Yes.

And for cyber-attacks, if you can switch to a different defense posture, then the attacker doesn’t gain anything and you learn a new zero-day exploit. And then you… This is called defense in depth. So, advanced warnings, threat intelligence to cyber-attack is like those seismic detectors for earthquakes.

And after that, maybe half a minute’s notice at most. For cyber-attack, that is exactly the same. But during those few precious seconds, for earthquake, if you can stop the elevator, then that prevents a lot of damage.

It is a very apt analogy because earthquake is not going to give you a lot of warnings either. They are not like typhoons, but if you work on zero trust architecture, just like in earthquake, we deploy the detection machines as close to the actual centers as possible, then ...