Matt Corallo 🟠

14.2K posts

Matt Corallo 🟠

Matt Corallo 🟠

@TheBlueMatt

10th known contributor to Bitcoin Core. Full-Time Open-Source Bitcoin+Lightning Projects @spiral_xyz. Open-Source Bitcoin for 15 years. Mostly reposting @soona.

NY/SF, usually Katılım Şubat 2011
264 Takip Edilen78.1K Takipçiler
Sabitlenmiş Tweet
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
It’s time for bitcoiners to step up and build. You don’t need to know anything about software development anymore, you just need to know how to write words! We have a golden opportunity to build out agentic payments based on open money, rather than letting agentic payments be captured by megacorps yet again. But we’re squandering it arguing about useless crap instead of building. Play with openclaw, give it a bitcoin wallet (moneydevkit makes this super easy! Also Lexe and phoenixd!), make it do things. If it fails to do what you want, go fix it! Have your agent build that bitcoin domain reseller, that bitcoin airline ticket reseller, or whatever it is you want! Bitcoin doesn’t just happen, it’s built. Join in.
calle@callebtc

USDC on base seems far more common for 402 payments now than Bitcoin. Recently, even Stripe joined the bandwagon. That’s a centralized stablecoin on a permissioned chain. Agents are starting to use fiat. It’s a huge loss, and in a race, many aren’t even aware that it exists. The scam coins are marching on, and even fiat is evolving. Where are the Bitcoin solutions that attract real users? Which other concept other than buying and selling Bitcoin has actually broken out of the bubble and made it to the mainstream? Bitcoin doesn’t just happen. These missing solutions need to be built by someone. Reject the “Bitcoin wins by hodling” narrative. The devs and entrepreneurs are what keep this project alive and keep marching forward. It’s not the scammy influencers, not the psychotic drama queens, the child-like infighting, or incompetent idiots dancing on the graves of word-class devs leaving Bitcoin. I hope the bear market flushes all that crap away, and we can get back to building stuff instead of tearing it down.

English
78
143
773
130.8K
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
We’re on a pretty similar page. This is what I put on nostr this morning: It’s still early, but it’s not at all the case that we cant learn anything from this and do better. If I were a hardware wallet manufacturer, I think my takeaways/changes I’d make are: 1) insecure fallbacks or hashing in weak entropy should be explicitly verboten. Unless it’s 96 bits or higher, do not include it! 2) Anything generating keys on an embedded device should have at least two random inputs and should check them with even basic randomness tests. This requires pulling lots of extra bits in before they’ve been whitened which isn’t possible on all hardware, but hardware that does whitening usually will do tests for you so make sure you’re checking it! 3. (Obviously) Everyone should be doing regular LLM reviews of software and firmware. 4. HWWs should have an explicit QA step where each input to the RNG is logged and the result of adding those inputs together to build a seed is done on a separate device with a reimplementation and compared against the device's seed. This should be done for each firmware release. I doubt almost any HWWs today do 2&4.
English
0
1
1
16
Jameson Lopp
Jameson Lopp@lopp·
1. Bug was non-obvious, not in the main code but in the build system. 2. Pretty sure Ledger DonJon reviewed coldcard several times over the years & missed it. 3. Most LLMs miss it due to not pulling in full submodule context. 4. Critical vulns have gone undetected for 10+ years in extremely popular code like SSL, SSH, & Linux. x.com/callebtc/statu…
calle@callebtc

it’s incredible that no security audit has found the coldcard deterministic RNG bug. it’s arguably the most important line of code in the firmware. they’ve done external security audits with real security researches right? …right?!

English
66
45
682
61.6K
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
@jessebabies @ChristopherA The point of defensive coding is so that bugs like this that are easy to miss in review don’t cause catastrophic failure. I posted some points I think hardware wallets should take to heart on nostr.
English
1
0
0
49
Christopher Allen
Christopher Allen@ChristopherA·
Re: Cold Card randomness. In 1994 Netscape released the first secure server. Verisign had a brief monopoly on certificates and would not issue a certificate unless the server had a basic security review. My company was the only one willing to offer these. Over HALF failed due to problems with randomness! It is so easy to make a mistake. Since then it has always been first on my list to get multiple eyes to review, not just my own.
English
6
5
96
7.2K
Pledditor
Pledditor@Pledditor·
NVK technically could have generated all possible seeds since there was so few, rugged everybody with good intentions, and tried to redistribute the funds using device IDs and email lists. But the obvious problem with that is he would go to jail for property theft, wire fraud, and various different cyber crimes. And the process to distribute funds back to CC users would be so messy, imperfect, and costly that I doubt in practice it would have worked out anywhere near as well as it sounds on paper.
English
1
0
1
352
Pledditor
Pledditor@Pledditor·
I've wondered this too. Once the devices shipped in 2021 preloaded with the bugged firmware, fate was determined on that day that large amounts of people were going to get rugged. There really couldn't have been much intervention to change this outcome. The coins were secured with such low entropy, it's basically the cheeto lock on the door meme. Once word got out of how weak they were, the coins were inevitably going to be exploited. If Coldcard had caught this issue early, the most ethical approach would have been to tell a few white lies: first, quietly patch the firmware, but in that firmware update, simultaneously release a brand new feature that actually is just a silly gimmick, but marketed as a landmark game changing innovation to HWW that you really can't live without. Get influencers to shill this "new feature" hard to entice FOMO, and allow that "new feature" act as the trojan horse covering up the real purpose of the firmware update: quietly patching over this disaster.
Michael Folkson 𓉱@michaelfolkson

If a security researcher had found this Coldcard bug before attacker it would’ve been really interesting. How the hell would Coldcard have told users to move funds without alerting potential attackers? I’m staggered no security researcher or dev at another HW wallet firm found it

English
8
2
69
8.1K
Clay Garrett
Clay Garrett@clay_garrett·
Sharing our initial findings on a reported vulnerability. Our recommendation is to continue to use your Bitkey normally. The reported vulnerability would require exceptional circumstances to exploit, and can only occur at a specific narrow time during inheritance setup. Even if an attacker was able to exploit this vulnerability (including exploiting TLS internet security), they would not have enough cryptographic material to access funds. This is Bitkey’s defense-in-depth in action. Our assessment is this presents no risk of remote drains or immediate funds loss. We appreciate @1440000bytes who reported this issue to us directly. We will share a more thorough technical report imminently, and follow that with a hosted space on X where the team will talk through the details with the community and answer any questions. We'll submit a patch to the mobile app to both stores today.
English
64
118
877
143.3K
Peter Van Valkenburgh
Peter Van Valkenburgh@valkenburgh·
Rather like chess engines again. Early gains came from finding obvious blunders. Now they're squeezing out fractional-pawn positional improvements twenty moves later. Maybe the security analogue is that the remaining bugs increasingly matter only under much narrower threat models, making them less relevant to the security of something like a hardware wallet. If that's right, it also suggests an interesting long-term argument for decentralized or least-authority systems. AI doesn't just make verification cheaper; it increases the relative value of systems with small attack surfaces, limited authority, and simple trust assumptions.
English
1
0
3
181
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
LLMs are very two-sided for security. On the one hand, long-latent bugs are finally surfacing. On the other hand, before you choose to use a product, you can get your own code review done for incredibly low cost*! In the short term it’s incredibly bad, in the long-term it democratizes code review greatly. “Don’t trust, verify” finally becomes an actual possibility for normal people! * of course LLMs do still make up issues when asked, push any such review to write PoCs to check their analysis!
English
16
20
161
19K
Matt Corallo 🟠 retweetledi
Rob Hamilton
Rob Hamilton@Rob1Ham·
We are likely entering the final hours of Cold Card MK4s, MK5s & Qs with "default" seed phrases generated on device from being safe on a single signature. Please reach out to anyone who you think may be impacted by this, every second counts.
English
14
83
343
26.7K
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
This is true, but as the complexity of the vulnerabilities increases, so does the complexity of exploit. For something simple like a hardware wallet, there’s kinda not all that many ways you can mass-exploit - either the randomness in the keys and signatures was sufficient or it wasn’t. Yes, the AIs may make the race for how to extract seeds from a physical device more competitive, but now you have to physically steal the device first. Maybe this is a problem for more complicated projects like an operating system or browser (more creative chaining of bugs to build a full browser escape or whatever), but even there there is (presumably) a finite number of bugs that can be exploited. As we get better at catching new ones before they’re added, maybe at some point AI is just good enough that we don’t ship new critical vulnerabilities anymore. Some will probably use this as a way to ship more complicated code that ends up trapped with new issues by each new model, but that doesn’t mean everyone will.
English
1
0
5
379
Peter Van Valkenburgh
Peter Van Valkenburgh@valkenburgh·
This got me thinking about chess as a model for AI and software security. One obvious story is that we'll get diminishing returns. Early AI gets dramatically better at finding bugs, but eventually improvements taper off and defenders catch up because everyone has access to roughly the same tools. The trouble is I'm not sure chess tells that story. If you measure chess by Elo, it's all relative. But if you measure something closer to absolute skill—average centipawn loss or distance from optimal play—engines don't seem to have hit a wall. They're still getting objectively better, even though they already look perfect to humans. The returns diminish, but they don't disappear. So maybe we shouldn't take comfort in the idea that AI capability itself will naturally plateau. The optimistic lesson from chess may be a different one. The biggest discontinuity wasn't Stockfish 15 to 16. It was the moment analysis became human+engine instead of just human. Overnight, every serious player gained access to stronger analysis than the world champion had a generation earlier. The economics of verification changed. Maybe software security is about to cross the same threshold. Right now AI is exposing decades of latent bugs. That's painful. But if "human+AI" becomes the default way software gets written and reviewed, then every commit, dependency update, and protocol change gets something like a superhuman code review for almost zero marginal cost. And as Matt points out, if the reviewer has to produce a working PoC rather than just speculate, "don't trust, verify" starts becoming a practical engineering workflow instead of an aspiration. So maybe the important question isn't whether AI capabilities eventually level off. Maybe it's whether we've already crossed the discontinuity where verification itself becomes cheap enough that defenders finally get to play a different game.
English
2
0
5
604
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
@4moonsettler @PeteClubSeven Sure, no rescue review is gonna be perfect, not LLM nor human. It’s still infinitely better for your average person wanting to evaluate a product vs trusting a few experts.
English
0
0
3
63
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
@4moonsettler I don’t think that’s necessarily the case, but even if it is, the LLM might be able to comment somewhat loosely on overall code quality and near-misses, which would be a good indication.
English
1
0
2
336
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
@4moonsettler @PeteClubSeven I dunno, they keep finding really useful stuff, and while, yes, probably half of what they find is nonsense, that’s still a pretty good hit rate given how much output they provide.
English
1
0
2
53
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
@SGBarbour @jyn_urso There will be more to shake out, but in the end we’re all gonna be better off for it.
Matt Corallo 🟠@TheBlueMatt

LLMs are very two-sided for security. On the one hand, long-latent bugs are finally surfacing. On the other hand, before you choose to use a product, you can get your own code review done for incredibly low cost*! In the short term it’s incredibly bad, in the long-term it democratizes code review greatly. “Don’t trust, verify” finally becomes an actual possibility for normal people! * of course LLMs do still make up issues when asked, push any such review to write PoCs to check their analysis!

English
1
0
4
861
Steve Barbour
Steve Barbour@SGBarbour·
Self custody and cold storage isnt over until Bitcoin itself is over. Bitcoin simply does not work without a strong self-sovereign, privacy-focused userbase. If everyone were to take the easy path and use State-surveilled custodial wallets, then it all ends with a literal bang 🔫🪖 (or worse - a whimper). Yes, Coldcard failed us. Many are rekt with many more to come. It is bloody out there, trust me I know... We are almost 20 years into this revolutionary technology, yet barely out of the first inning. Enough with the infighting and petty bullshit FUDslinging. Let's pick each other up, there's a lot more work to do!!
English
11
8
91
3.8K
Matt Corallo 🟠 retweetledi
Kevin Loaec 🧙‍♂️🐟
Kevin Loaec 🧙‍♂️🐟@KLoaec·
It's happening. Mk4, Mk5, Q are now actively drained. Breaking an Mk4 is HARDER than breaking a weak passphrase, so your Mk3 "passphrase protected" are at immediate risk (if less or around 32 bits entropy).
Tomer Strolight@TomerStrolight

An acquaintance of mine intentionally left a small amount of bitcoin in a ColdCard MK4 RNG created seed phrase (that was originally a "duress" wallet) to see when it might get swept. Last night it got swept to bc1qzm5pauxyv7t7vqstzpumqcn066wfjsmev34mf3. So at this point any RNG generated seedphrase on any ColdCard product is under active attack. Move quickly if you're exposed.

English
83
234
1K
664.1K
Matt Corallo 🟠 retweetledi
Rob Hamilton
Rob Hamilton@Rob1Ham·
I am urging engineers who are in the Bitcoin ecosystem, use @OpenRouter or @opencode to use @Kimi_Moonshot K3 on any software you run which interacts with bitcoin in any way internally and any public repos you use. Other models CAN find issues, but K3 will one shot full vulnerability reports to get an extra set of eyes for security checks. Some of it will be slop/overstated/wrong, but I've been scanning open source repos and finding things I'll be passing along to maintainers. Kimi K3 came out as open weights on Monday, and I don't think that is a coincidence as it relates to the COLDCARD issues unfolding.
English
36
140
830
206.5K
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
@Igbeauxx @Leishman A fairly common vendor and probably quite common for River’s userbase. Delaying auto-withdraw by a few days (at worst, for folks who can’t click a link) seems a pretty low price.
English
0
0
3
77
Alexander Leishman 🇺🇸
Alexander Leishman 🇺🇸@Leishman·
If you have auto-withdraws enabled on River and it was sending coins to a ColdCard please disable the auto-withdraw and point it to a new destination if you have a safe secondary self-custody system.
English
2
31
288
14.4K
Tom Turney
Tom Turney@no_stp_on_snek·
284B DeepSeek-V4-Flash-0731 at 2.88 bits/weight. 95 GiB (GGUF) doing some local testing while i start the upload
English
5
0
52
5.8K
Bryan Bishop
Bryan Bishop@kanzure·
@raw_avocado @TheBlueMatt The norm should be that people who want to use AI on an open source project should donate their tokens instead of donating their outputs.
English
1
0
1
111
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
@AchimWar Sure but you also aren’t going to be exposed to 2029 vulnerabilities until 2029 :)
English
0
0
0
156
Matt Corallo 🟠
Matt Corallo 🟠@TheBlueMatt·
@sorukumar I’m not suggesting everyone has to, but the fact that bugs are getting so shallow means it only a handful of people do this new vulns will be found quickly!
English
0
0
2
194
Saurabh
Saurabh@sorukumar·
I see the point. but the implicit assumption in the long-term view feels a bit shaky. expecting everyone to do everything, is one of the main reasons UX stays so bad. here the product builder is just offloading their lazy work onto users. even post-AI, why would we expect a doc to code-review the hw they’re planning to use?
English
1
0
1
287

Keşfet

@lopp @jessebabies @ChristopherA @Pledditor @ZLOK @clay_garrett @1440000bytes @valkenburgh