YT YouTube2Text

YouTube2Text → Transcripts → Will AI Break Crypto, or Make It Stronger? - TOKEN2049 Singapore 2026

Will AI Break Crypto, or Make It Stronger? - TOKEN2049 Singapore 2026 — Transcript

by TOKEN2049 · 4,444 words · 639 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:08Well, thank you everybody for coming.
0:10Uh, excited for this panel. Um,
0:13obviously there is a massive tsunami
0:16coming uh in the technology field which
0:19is AI. Um, it kind of feels like crypto
0:21in some sense where it still feels very
0:24early but kind of unknown. Um and so
0:28great to have these esteemed guests uh
0:30join me. Maybe to start we can just do a
0:32quick round of intros and then jump into
0:35things.
0:36>> Sure. Hey everyone, my name is Cosmo. Uh
0:38I lead investment strategy at Panta
0:40Capital. Panta is a tech investment
0:42firm. We have early stage venture all
0:44the way through public markets
0:46practices. My name is Drake Evans. I'm
0:48the CTO and co-founder of Agora. Um
0:50Agora provides a single and final entry
0:52point for internet native money. Uh we
0:55have a stable coin AUSD. We also do
0:57white label issuance. Um and in soon uh
1:00with the finalization of our bank
1:02charter, custody payments, US dollar
1:04bank accounts. Um we also do wallets.
1:06Really the one-stop shop.
1:08>> Awesome. Um hi, my name is Jonathan. I'm
1:11co-founder and CEO of Chain Analysis. We
1:14provide all of the intelligence
1:16uh for the industry and regulators and
1:19government agencies that want to be able
1:21to actually protect crypto and make it
1:24stronger.
1:24>> Perfect. And I'm Logan. Uh I lead
1:27Frictionless Capital uh early stage
1:29venture fund. Um maybe starting with
1:31you, Jonathan. Uh kind of diving deep
1:35onto onchain. We were talking right
1:37before we walked on stage on the
1:39cryption side. Um it's a little bit of a
1:41cat mouse game. um between AI kind of
1:45trying to poke holes in code versus what
1:48we have with smart contracts. What's
1:50kind of your current read of the market?
1:52>> I think one of the things that we start
1:55to see is that you know AI is clearly
1:57being used in exploits that are being
1:59leveraged uh on you know attacking
2:02protocols and chains and you know even
2:05wallets. And I think that you know the
2:07difficult thing that we have to do as
2:09defenders is that there is always going
2:12to be human in the loop currently on
2:15defense
2:16whereas there is not necessarily human
2:19in the loop on attack. And when you have
2:22that type of asymmetry, you have issues
2:24on like what speed you can operate at.
2:28And I think that this is where you start
2:30to see problems occurring um in in the
2:34in the sector. That being said, the
2:37biggest, you know, thefts of this year
2:39are still not like agent swarms hacking
2:43into people's stuff. It's actually just
2:45like getting the basic cyber security
2:47stuff right. And I think the you know
2:49private key compromise and other things
2:51that we've seen you know are just
2:53traditional cyber security failures
2:56rather than you know completely brand
2:58new exploits.
3:00>> And Drake we were talking earlier again
3:03about Justin Drake's uh Twitter post uh
3:06if you have not seen it uh got fairly
3:09viral on Twitter and asks about kind of
3:11impending doom of AI on crypto markets
3:15and potentially breaking cryptography.
3:17And you also had some very interesting
3:19thoughts here. Yeah, I mean I think if
3:21you've been you know looking at Twitter
3:22the last 24 hours there's been a lot of
3:24discussion about the you know a bunch of
3:27old old math proofs uh or math problems
3:30being solved or you know these sort of
3:33assumptions about the barriers or these
3:35asmmptotes being broken in you know what
3:39people are calling like in in an
3:42inelegant solution right I think humans
3:44have a tendency to find elegant
3:45solutions because of small context
3:48window and other
3:49Uh and so then the the thesis goes okay
3:52you know the the big labs have posted
3:55100 uh math proofs that they broke in
3:57the last few months zero of them are
3:59related to cryptography that's a bit
4:01suspicious so you know if there's a
4:03nonzero chance that there is a inelegant
4:06solution to significantly shorten the
4:08cycle of breaking what I would consider
4:11most of the you know asymmetric
4:13cryptography algorithms as elegant you
4:16know we should be preparing for that I
4:18think there's been a bit of a
4:21the counterargument of course is that AI
4:23will help us find better algorithms for
4:25the future [snorts]
4:27>> and Cosmo Panta has been one of the OG
4:30OG crypto funds. You guys have seen a
4:34lot of high highs, some low lows. What's
4:36kind of your current market take more
4:38from an investor standpoint about kind
4:40of maybe the shifting landscape as
4:43there's new paradigms? Uh maybe this is
4:46not the first because Panta's been
4:49around for so long, but how are you kind
4:50of thinking about it from the investor
4:51view?
4:53>> Well, I think the biggest paradigm is
4:54clearly the impact that AI has on any
4:57industry. And so as uh as I think as any
5:01unique human, I'd encourage you to think
5:03about how you and your specific skill
5:05set or toolkit can be used to accelerate
5:08what is one of the biggest tsunamis as
5:10you describe Logan, which is, you know,
5:11the influx of AI innovation and its
5:13penetration of society. And so us as
5:15blockchain investors naturally then we
5:18are thinking about how does blockchain
5:19play a role in accelerating AI
5:22innovation or on the flip side how does
5:24AI impact blockchain adoption one of
5:26which is maybe breaking crypto but the
5:28other is accelerating crypto and you
5:31know I I think this is you know maybe a
5:33controversial take and I'm sure there's
5:35holes with this but I do believe we're
5:37nearing the end of smart contract hacks
5:39like the era of smart contract hacks I
5:40think will be over in the next 3 to 6
5:42months because anything that can be
5:45hacked will have been hacked by then and
5:47that anything going forward will have
5:48been you know fully penetration tested
5:50before it's been released and so you
5:53know the optimist is that we're almost
5:54done with DeFi hacks now that doesn't
5:56stop anything Jonathan said about human
5:58hacks right which is really the core of
6:00the problem um but like cryp yeah yeah
6:02so crypto's
6:03>> maybe not the problem in the future it's
6:05really humans remain the problem
6:06>> I think that's absolutely right
6:08especially you know if you look at sort
6:09of traditional software versus smart
6:12contracts surface area for smart
6:14contracts is just generally
6:15significantly lower. And I'm, you know,
6:18very optimistic that the defensive tools
6:20are getting better and better to the
6:21point that you can have as close as 100%
6:24certainty a smart contract secure. Um,
6:27traditional software is a bit harder one
6:29because it's constantly changing and
6:31there's typically a lot more complexity
6:32and and uh service area.
6:35>> I'll take the other side. Okay. I mean,
6:37I think um no, I think that like there
6:40is no such thing as like secure like
6:42completely secure software. And I think
6:44that we're at this point where not all
6:46of those languages are necessarily sec
6:49like able to produce uh secure smart
6:53contracts. And I think that at the end
6:55of the day, you know, whether it's
6:56agents writing them or whether it's
6:58humans writing them, there's there's
6:59going to be failure modes inside of
7:01that. I think that the problem may
7:04change and I think you're both right the
7:07defense against these things has got a
7:09lot better. It you know the operations
7:11in projects is getting a lot more
7:14sophisticated. Um and yes like lots of
7:18things that were going to get hacked
7:20have been hacked. But I think that you
7:22know fundamentally I don't I don't think
7:24we can live in a world where we think
7:26that there are going to be fundamentally
7:28secure systems that everyone can just
7:29rely on. So assuming that maybe we
7:33aren't at the end of finding all the
7:36bugs and that there is still going to be
7:40some potential hacks or bugs that are
7:42are exploited with AI over time. What's
7:45the best way to try to protect about
7:47that? Is it formal verification? Is it
7:50kind of AI defense? Where do you think
7:52that kind of leads us? Yeah, I think
7:54that maybe one of the things that I
7:56think you know will be necessary, I
7:59think there will be more formal
8:01verification methods uh in general and I
8:03think that AI actually will lower the
8:06barrier to entry for some of those
8:08programming languages that have
8:09historically been very difficult to get
8:12people to adopt in even in the crypto
8:15industry. Um, but actually I think that
8:18the biggest challenge that we're going
8:19to have as an industry is actually not
8:21in the smart contract vulnerability
8:23landscape, much more in like fishing and
8:26scams. Like this is this is something
8:28that I I don't think is even like a
8:30crypto problem. Like this is just a
8:32global problem. Everyone here has like
8:35experienced unbelievably good video,
8:40audio,
8:41textbased AI that is so believable that
8:46will morph someone's ability to
8:48understand whether they're being scammed
8:50or not. Uh, and I think that the best
8:54form of defense to the question is
8:56actually offense. And I think that you
8:59can't think that we're going to train
9:01humans to detect AI scamming them from
9:05things. You're going to have to go meet
9:08scammers at scale using agents
9:11pretending to be victims to be able to
9:14actually uncover the financial
9:16infrastructure that they're running on
9:18in order to best actually address this
9:20problem. And this is what we've been
9:22doing at Ch analysis for the last two
9:23years is like we've been building agents
9:26forms effectively to go actually and
9:29talk to scammers at scale to be able to
9:32solicit all of their infrastructure that
9:34they're running on. And that's like the
9:36only way that you can get like real
9:38preventative,
9:40you know, care against these types of
9:42threats.
9:44Dossing scammers.
9:46>> Not just DDoS, but actually like I think
9:48that this is the thing is like it used
9:50to be the case that you could like waste
9:52a scammer's time and then they wouldn't
9:54be able to call Cosmo, but like now like
9:57they've got infinite capacity. So what
9:59you have to go and get is their like
10:03financial infrastructure that they're
10:05relying on which is scarce as a
10:07resource. Mhm.
10:09>> Cosmo from again maybe more the investor
10:12lens, DeFi was really one of the core
10:14primitives and continues to be one of
10:16the core primitives within blockchains
10:18and kind of these capital markets that
10:21we're building. Does it change kind of
10:25your view on the investibility of any of
10:28these assets or again do you think all
10:30these DeFi bugs just get solved within
10:32the next 6 to 12 months? I would say it
10:35definitely gives us some pause about how
10:39we should be allocating capital in
10:40today's world. Um the
10:45there there's no doubt that
10:48that a lot of this stuff is you know in
10:51not fully secure as as has been
10:53described. I think there's a you know
10:55there's a risk tolerance that we're okay
10:57with and there's by the way like a lot
10:59of things can survive hacks. Um I mean
11:02almost all of the centralized entities
11:03have survived tax, right? And so there's
11:06a little bit of like the market kind of
11:07sees through it. Uh because at the end
11:08of the day a business is more than just
11:10like hasn't been secure 100% of the
11:12time. It's more that doesn't provide a
11:14consistent reliable experience than like
11:1659s maybe. Um and so you know there is
11:19risk. There's definitely increased risk
11:20and we're definitely more
11:23uh just more more cautious more curious
11:25around like net new protocols or things
11:28that haven't been as battle tested. Uh
11:30but you know there's always stuff to do
11:32and you can't be can't be scatter your
11:33own shadow. And Drake, we were also
11:36talking about kind of the lendiness of
11:38some protocols is really like you can
11:40stress test them as much as you want,
11:41but really you just kind of need time in
11:43the market, but there's also kind of
11:46this like third factor that people have
11:48been talking about on the timeline,
11:50which is quantum. And kind of you not
11:53only do you have AI, but potentially you
11:55have quantum coming up. And what are
11:57your kind of thoughts about quantum as
11:59well? I I think it's probably very
12:02related to what we were talking about
12:03earlier and just that AI and and this is
12:05like I think a general thing that will
12:07affect crypto and all industries is AI
12:09is going to accelerate a lot of existing
12:12trends. I think it will accelerate
12:14tokenization. It'll also accelerate our
12:16path to breaking quantum which I think
12:18most people had accepted as something
12:21that would happen in the next decade or
12:23two um even conservatively. And so, you
12:26know, I think it's funny, Vitalic posted
12:29today about kind of like how all of the
12:32plans they've been implementing over the
12:33last few years like play into this kind
12:35of roadmap. And I think um you know,
12:37teams who have that forward-looking kind
12:39of mindset and and organizations that
12:42sort of operate on the assumption that
12:44things will move faster than they
12:46expected will do very well. Um if you
12:49look at even just AI over the last year
12:52um you've seen this just like what feels
12:54like an exponential growth in capability
12:56but there were people in 2023 saying
12:59this is you know you've seen that the
13:01meme graph right where uh it feels like
13:04it's just it's like the intelligence of
13:05an insect and then a dog and then a baby
13:08and in a blink of an eye it'll be you
13:10know better than a human. And so I think
13:11we're seeing that play out and and
13:13making the bet that timelines will
13:15compress, innovation will will go faster
13:17is is the only way to stay ahead of it.
13:20>> So this panel has been a little bit doom
13:22and gloom uh in the sense that AI can
13:26really kind of throw a wrench in things.
13:28But maybe an open question to the panel
13:30is what are the things that are getting
13:31you guys excited now? Um crypto markets
13:34have matured. Uh we've run a lot of
13:36experimentations. Obviously, you guys
13:39each are still in the industry. I'm
13:41assuming because there is some optimism.
13:43What are the things that currently
13:44excite you?
13:46>> I I think it's it's really just the
13:48acceleration of all of these existing
13:49trends. I think that AI will accelerate
13:52tokenization. And I think the big thing
13:54is AI is you know with there's a lot of
13:56uncertainty but it's definitely lowering
13:59the barriers to doing more stuff broadly
14:03whether that's starting a new business
14:05uh you know in some cases attacking
14:07software but like in general like people
14:10are going to be doing more stuff that
14:11means more innovation that means more
14:12creativity that means uh you know we're
14:15going to see really cool stuff I think
14:16over the next few years. I think that
14:18like from from our perspective firstly
14:22like we're using AI now at scale to like
14:25you know tackle some of the big threats
14:27to the industry and like it's actually
14:29helped us get ahead of some of those
14:31threats you know particularly on the
14:33scam side but also you know the various
14:36different you know types of elicit
14:37activity that we that we track at chain
14:39analysis like we're able to leverage it
14:42to be able to protect the industry at
14:44greater scale. I think the next chapter
14:46of this is actually helping all of our
14:48customers leverage AI to be able to
14:50perform investigations to do compliance
14:53more efficiently to really you know save
14:56on you know the time and expense that
14:58goes into you know running you know
15:00operations uh at different uh businesses
15:03so that people can be focused on like
15:05growing their business and and doing uh
15:07better for their customers. I think that
15:10you know AI will shorten the latency
15:12time for response for customers. is it's
15:15going to increase you know how
15:16accessible crypto products are in
15:18general for everyone. I think that the
15:21things that we're seeing that are
15:22exciting at chain analysis is you know
15:25coming to this region there is just so
15:27much more happening on payments in uh
15:30APAC than you know in other parts of the
15:33world. There's lots of stuff happening
15:35for the real goods and services that are
15:37moving uh based on stable coin rails.
15:40There is also like a a large
15:42intersection of you know AI and crypto
15:45in you know the inference market and uh
15:49seeing you know inference being paid for
15:52uh using crypto is is something that we
15:54start to see a lot of at chain analysis
15:57and you know I would say that you know
15:59it's it's a great moment for people to
16:01innovate on new types of financial
16:04services based on stablecoin banks that
16:06we heard from the previous panel. I
16:08think like this is something that you
16:10know has exploded over the last you know
16:13year since Genius uh in the states and I
16:16think that you know that's only
16:18something that we'll we'll continue to
16:19accelerate. Um so I think that I'm like
16:22very bullish on you know the ways in
16:25which we're now tapping into real
16:27economic activity running on blockchain
16:30rails. Um it's what you know excites us
16:33the most at chain analysis is is seeing
16:35those flows. Um and actually like I
16:38think that you know it's it's an awesome
16:40moment to be a builder. Like it's an
16:42awesome moment to uh be able to
16:44experiment and like that experimentation
16:47is what crypto is actually fundamentally
16:49all about.
16:50>> I think the riskreward for investing in
16:52blockchain protocols today is the best
16:55that it has been in many many years. And
16:57and the reason I say that is precisely
17:00because I think AI is going to be such a
17:02huge accelerator of onchain activity. If
17:06you think and largely because of agentic
17:08activity when you think about like what
17:10agents are agents will want to be able
17:13to transact with each other. They they
17:15want to uh and when they transact with
17:18each other you know they'll have to find
17:19some sort of transaction rails and they
17:21they have option A option B. Option A is
17:24walk into a physical branch, open,
17:26present two forms of ID, wait a few days
17:28to get a bank account, and then have the
17:30privilege of sending money 5 days a
17:31week, 8 hours a day, or spin a
17:34blockchain wallet that's 24/7, global,
17:36cheap, fraction of a penny anywhere
17:39instantly, they're obviously going to
17:41choose option B. And so as we see to the
17:44extent you believe at all in the growth
17:46of AI in the growth of the agentic
17:48economy in the growth of agentic
17:50intermediate econ commerce like it's
17:52obviously going to accelerate everything
17:54across blockchain and that is such a
17:56massive tailwind and at the same time
17:57you know you see there's plenty of ways
17:59to invest in AI we see all in AI stocks
18:01and AI related things go up a ton
18:03definitely not saying they are
18:04necessarily overvalued by any means but
18:06they've gone up a lot relative to
18:09blockchain which is down 40 50% from the
18:11highs.
18:12even though they're riding along the
18:13same secular tailwinds. And so to me
18:15that seems like an incredible incredible
18:17riskreward.
18:18>> Maybe back to you Cosmo. And I think
18:21right now there's a lot of people
18:23probably in the audience are trying to
18:24learn about what is the best way to
18:26express going long crypto, going long
18:29AI, the possible intersection. Is it on
18:32the agent side? Um how do you kind of
18:36see that? Well, this is where like you
18:38have to drop down from generalisms to
18:40actual ideas that are interesting
18:41because there are a lot plenty of ideas
18:43that are interesting, but there it's
18:45execute name of the game is an
18:46execution. I mean the guys at ch
18:48analysis and agor have done an excellent
18:50job in execution. You should check out
18:51both their products. um not a plug and
18:54uh I I'd say like within within
18:57blockchain protocols some of the
18:59protocols we find most exciting are
19:01anything across uh you know deepin
19:04protocols I think are going to be very
19:06interesting in a world of AI it's very
19:07clear that some of the bottlenecks to to
19:10AI you know adoption are energy compute
19:14data there are plenty of deep networks
19:16across energy compute data right there's
19:18grass doing amazing things across data
19:20there's glow and others doing amazing
19:21things across energy
19:23compute. There's all sorts of ways that
19:25people are thinking about distributed
19:26inference and distributed training.
19:28Distributed training may be tougher, but
19:29distributed inference. And so that's
19:31like super exciting. Or whether across
19:33identity we just led around in uh in
19:35Worldcoin, which is, you know, been this
19:38thing that's sort of been kind of a
19:39crazy dream and a crazy promise for a
19:41long time. But this concept of proof of
19:43human is actually really relevant today.
19:45The idea that you and I are going to do
19:47more online and that when we're in the
19:49digital world, the desire to know who is
19:51truly human will only increase over time
19:52and that identifier will become very
19:55valuable over time is is massive.
19:57>> And Drake, how are you guys
19:59incorporating kind of AI within Agora?
20:02>> I mean, I think I think there's sort of
20:04two ways to think about it. The first is
20:06how are we using AI internally and then
20:08the second is what do we believe about
20:09what AI will do to our customer base.
20:12Um, I think, you know, on the
20:14customer-based side, we believe that
20:16you're going to have to compete. You
20:18know, consumers are no longer sticky in
20:20in the same sense, right? There's been a
20:21lot of talk about this with bank
20:22deposits, but but it's really you're
20:24going to have to have the best product
20:26and the best economics and have the
20:28aggregation and lower that friction. Um,
20:31you know, we were just talking about
20:32how, you know, choosing a wallet over a
20:34bank is so easy partially because the
20:36friction to value is so fast, right?
20:39There's so little friction to get the
20:40value. I think so that's going to change
20:41how people think about consuming
20:43products. And so we're focused very much
20:46on how do we lower the friction to first
20:48sort of value and then how do we provide
20:51those base building blocks because
20:52that's the other piece that is going to
20:53change is that people are going to start
20:55building they want access to the metal
20:57sooner and sooner and the deeper
20:59infrastructure because now you know this
21:01whole middle ground where you paid a SAS
21:03product to kind of combine a bunch of
21:05stuff for you. um you can do the
21:07combinations yourself and so now you
21:08want the Legos and then on the internal
21:10side I think you know we view every
21:12single human decision as an opportunity
21:17right anytime that a human needs to do
21:18something and is a process like that's
21:21an opportunity to build a custom harness
21:23to help them with their workflow that
21:26over time you know starts off providing
21:27recommendations then you know at high
21:30confidence is automating things and so
21:32you know the first place we really
21:34pushed on this is in our onboarding
21:36where you know when when uh our
21:39onboarding team wakes up in the morning
21:40they have you know a brief of everyone
21:42who's applied what potential next steps
21:44might be what things that might flag
21:46that and then they can sort of rate
21:48those those um uh and then you collect
21:51that data that becomes training data
21:52eventually it's large enough you can do
21:54post- training so really and and that's
21:56true for security compliance for legal
21:59like everyone in our company wants a
22:01custom harness that wraps their workflow
22:03that becomes a structured input but for
22:05what they're doing that can then
22:07automate it.
22:09>> And Jonathan, you you talked earlier
22:10about the agent swarm. I I think it's a
22:12very interesting idea just picturing it
22:14in my mind as you have these giant agent
22:17battles going back and forth and you
22:19you're putting up your 10,000 agents
22:21defending the wall and making sure smart
22:24contracts don't get exploited. maybe
22:26like dive slightly deeper into that
22:28because I do think it's a very
22:29interesting kind of mental model of
22:31defending the smart contracts and you
22:33guys are really very close to the metal
22:36at what you're doing.
22:37>> Yeah, I mean maybe um to give you a
22:39sense of like what we look at on a daily
22:42basis. I mean we look at not just smart
22:44contracts but we also look at like every
22:46website that's registered in the world
22:48every day and you know at a scale where
22:51we're trying to map out you know what
22:53are the transactions that are targeting
22:56you know these different smart contracts
22:58or you know how are these websites
22:59targeting you know investors or or
23:02people to you know try and get you know
23:04scams off the ground and all of that is
23:07like how do you respond to that
23:09instantaneously? How do you how do you
23:11capture evidence that someone can then
23:14rely on? And I think that AI is there
23:17to, you know, capture what humans were
23:19doing before and you can store it as
23:22evidence and serve it to people as
23:24facts. And I think that this is where,
23:26you know, we get away from some of the
23:29probabilistic
23:32stochastic models out there and you
23:35still can like use it as a tool in the
23:37toolkit but produce like really reliable
23:40evidence. And I think that this is where
23:42um you know the future of defenders
23:44really happens is like translating you
23:47know all of this like agentic work which
23:50is probabilistic in nature to like
23:52hardcore facts that you can act on. And
23:54that's that's a lot of what we've been
23:56spending the last couple of years
23:57thinking deeply about.
23:59>> Awesome. Well, I think we're at time so
24:01please uh help me in giving all these
24:03gentlemen a round of applause.

About this transcript

This page contains the full transcript of Will AI Break Crypto, or Make It Stronger? - TOKEN2049 Singapore 2026 by TOKEN2049, generated from the public captions YouTube serves with the video. The transcript has 4,444 words across 639 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.