Sam Noble (@sam_noble) header
Sam Noble (@sam_noble) profile picture

Fractional CTO · Dubai · SG entity. Scaling fires, 0→1. Ex-ZebPay ($1 bn/day), Veera (6M users). Chromium, wallets, token + agent custody. [email protected]

Dubai, United Arab Emirates
Joined June 2009
Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)Photo by Sam Noble (@sam_noble)
I've been writing production software for 20+ years. • Mined Bitcoin in 2009 and lost it when the laptop was stolen. • Helped launch chains. • Took an exchange from $200k/day to $1bn/day. • Put a self-custodial wallet and a rewards engine inside a Chromium browser that reached 6M users. I now take the CTO seat by the month. The parts that actually transfer:
3
4
748
🚨 We're releasing an emergency update to address nginx CVE-2026-123007, CVE-2026-123634, CVE-2026-123633, CVE-2026-121810, CVE-2026-113425. We recommend updating your BTCPay Server as soon as possible.
15
143
28
282
40,966
Yikes. Haven't dug into the details, but better safe than sorry. This appears to impact WhatsApp and Telegram desktop apps. Disable auto download of all media types in both apps: - WhatsApp -> Settings -> Storage and data -> Media auto-download -> Set all to off - Telegram -> Settings -> Data and Storage -> AUTOMATIC MEDIA DOWNLOAD -> Auto-Download Media -> Off
Last night I was hacked. The hack was delivered via WhatsApp payload. Downloaded automatically and gave access and or computer use to the attacker. They waited 5 days, following my tweets on X of short positions and closed everything and withdrew directly in one transaction from Hyperliquid > Metamask > Tornado Cash. The attacker waited patiently and closed shorts as the market got to 68k and withdrew everything. Clearly a crypto native attacker. They also gained access to my metamask, email, websites, and all of my passwords including X. It’s the most brutal thing as I kept everything over the past few days in crypto and on-chain off Coinbase/Kraken. I was planning to move funds back to my Revolut before Christmas time. Every single penny in dollars was taken, apart from $800. It’s gone, and in the process of being washed. Truly the most devastating day I’ve had in my life. Completely shocked by it all to be honest. I will recover, but this was very sophisticated tech, it caught me completely off guard. The timing was impeccable and it’s obviously been planned for a while. Stay safe everyone, will keep you updated. I fear the worst is to come and this incident has set me back many years of savings, time and hard work. Brutal.
1
1
99
This doesn't look like a zero-click issue and requires you to open / preview the payload, however having it on your machine is a liability. As always, be very careful what you open.
17
I seriously hope this is ragebait dressed up as wisdom, but in case it's not and someone takes the post seriously, here's the reality: This view is two years out of date. It reads like someone who hasn't ever scaled a team past the point where process and judgment actually matter, and who isn't using frontier models on anything more complex than a food delivery app. I've scaled engineering orgs from single digits to tens to hundreds, and spent the better part of 4.5 years hands-on with a Chromium fork, one of the nastiest, most complex codebases in existence. The delta between what I could ship two years ago and what I can ship today is not incremental. It's an order of magnitude. What used to take weeks now takes hours. Software fundamentals are well-defined, well-understood, and extremely well-represented in training data. Modern models and the agent harnesses around them adhere to them more consistently than most engineers. Writing code is now a straightforward transformation from high-level specification into implementation (which is what we always really wanted). Its laborious, error-prone translation work. Grunt work. If you think you still need to personally know why an architecture is appropriate, why an API might fail, why a database won't scale, or why something is insecure, look at the volume of CVEs being raised right now. A huge share of them are old bugs: flaws that sat undetected for years in code written and reviewed by humans who supposedly knew the fundamentals, now being surfaced at record pace because models are being pointed at decades of legacy code. Just look at the Microsoft 365 outage that's been burning for the last day. Tell me what about any of that suggests humans actually understand these systems at the depth you're claiming. The failure modes are the same ones we've had for decades. Agents didn't create that gap. They're the ones finally exposing it. Your points are table stakes. If that's the ceiling of what you spend your days doing, you should be worried. Those are the exact questions a competent agent plus a person who can clearly state constraints and acceptance criteria can already answer faster and more thoroughly than most senior engineers. The scarce skill is no longer "knowing" those things in the abstract. It's being able to apply them under real constraints, at speed, while the cost of exploration has collapsed. I had a version of this conversation recently with execs of a huge product company. The people still saying "agents can't replace engineers" are the ones who checked out in 2023. I might have agreed with them three years ago. Not today. The people who are still manually writing most of their code "for understanding" are the modern equivalent of insisting on punch cards because "you need to know the machine." If this is your actual model of the next 3–5 years, you've already lost.
My prediction for software engineering over the next 3 to 5 years. Computer science and software engineering fundamentals will become far more valuable than knowing the latest frameworks. If AI can write code in almost any language, knowing the nuance of another framework becomes less of a competitive advantage. But understanding algorithms, data structures, networking, operating systems, databases, architecture and security? That's a different story. You still need to know: - Why a particular architecture is appropriate. - Why an API might fail. - Why a database won't scale. - Why an apparently working solution might be insecure. A coding agent can build a solution. But you need the knowledge to evaluate the trade-offs and recognise when it's wrong. So, I think we'll see a renewed emphasis on the fundamentals of computer science and software engineering. TL/DR: Learning how computers and software systems work will matter more than memorising how to use them. What do you think?
1
118
This is really concerning from a privacy angle, especially at the SDK level which so many third-party wallets will rely on. Even "benign" metadata that may eventually surface on-chain (e.g. chain, recipient/contract, device model + firmware, app versions, blind-sign status) tracked across every signature lets someone reconstruct a detailed picture of your activity. Combine that with IP logs or other signals and the story gets richer. This really reminds me of Google logging IPs via Safe Browsing API calls under a safety pretext. You don’t know their intentions, retention policies, or what happens if (more like when these days) there’s a breach or policy change. Better to limit the flow of information by default. Privacy should be opt-in, not something one has to to discover and override after the fact and then re-check on every SDK bump.
If your wallet of choice isn’t @ambire and it runs Ledger’s latest signing kit, chances are every signature you make gets reported back to Ledger. I noticed it early while bumping Ledger packages to debug an issue! No affected version ever reached Аmbire production, and we’re shipping it disabled by default, with an opt-in toggle in the privacy settings 🛡️ However, the devil was that it was sneakily enabled by default in the Ledger SDK since their ledgerhq/context-module v1.16... So right after the lib update - every transaction you sign with a Ledger device started sending to blind-signing.api.ledger.com the network, the contract or recipient address, device model and firmware, app versions and whether you signed blind. I'd argue that privacy should be the default, @Ledger 💪 Happy to see more and more people becoming aware of the privacy implications and picking Ambire for our ongoing privacy focus. Catching things like this 👀 is part of the job for us!
2
2
164
We’re becoming too centralised on Cloudflare. Don’t get me wrong, I run a ton of infra on Cloudflare, they’re a great provider and it’s a powerful tool in the box, but just because you can run everything in one place doesn’t mean you should (spoiler: you really shouldn’t). Zooming out, I’ve run WireGuard (the underlying transport layer for Tailscale) since 2019. ~4000 lines of kernel C, auditable in a weekend, and almost no overhead. It was so efficient and secure that a lot of the EOS validator network peered using wg tunnels with no impact to our 500ms block times. Tailscale kept all of that and added the two things WireGuard (rightly) refuses to do: hand out keys and punch through NAT. It does not stand in the middle of the tunnels, peers still talk directly. If Tailscale’s control plane dies, already-peered nodes keep talking; you just can’t enrol a new one. Mesh is a different breed. It’s MASQUE by default, and every packet routes through Cloudflare servers. Your laptop talking to your phone on the same desk, same LAN, goes via Cloudflare. Cloudflare down means your VPN is gone. Four outages in twelve months: Nov, Dec, Feb, and on 20 Oct the Zero Trust gateway itself for three hours, dashboard included. That’s a dependency I’d rather avoid. Rule: the route you use to fix things must not share a failure domain with the thing that broke.
Je tourne toute ma stack derrière Tailscale. Serveurs, agents, labo perso. Ça marche, je touche plus. Là Cloudflare sort Mesh en beta. Même job en surface : IP privée entre mes machines, chiffrement post-quantique, accès sans VPN à la main. 50 nœuds et 50 users gratuits. Et comme j’ai déjà Cloudflare partout, Zero Trust, Workers, le truc est déjà dans mon compte. Le détail qui change vraiment : tout le trafic passe par le réseau Cloudflare. Pas de peer-to-peer comme Tailscale. Chez eux, c’est un choix, pour coller les policies Gateway et Access dessus, et pour que tes agents Workers puissent joindre ton labo privé sans exposer un port. J’ai encore aucun nœud. Tailscale, ça reste mon default. Mais un outil de moins, intégré à ce que je paie déjà, ça mérite un vrai test.
2
179
Great list, but with everything that's happening there's so much more that needs to be done, especially if you're an engineer (i.e. pulling Github repos) or in web3. If you're on a Mac, this is the absolute bare minimum: 1. Lockdown Mode on. 2. Daily driver is a standard user. Admin is a separate account you log into to install things and nothing else. Never type the admin password from the daily session. 3. No live runtimes on the daily profile. That means python, node, java, pip, npm. Use a shim that alerts when something calls them. Real ones live in a container or a VM, under a different user. 4. Objective-See, most of it. LuLu for outbound, BlockBlock for persistence, KnockKnock to see what's already persisted, OverSight for the camera and mic. Free, open source, and the author reads the same malware you're worried about. 5. "Privacy & Security" settings audit, once a week. Full Disk Access, Accessibility, Screen Recording. Every RAT I've seen on a Mac needed one of those three and got it from a tired human clicking Allow. If you don't know why an app is in there, it isn't any more. 6. Homebrew casks are the supply chain. Formulae build from source; casks pull whatever binary the recipe points at. Read the recipe before you install, and never one with sha256 :no_check. 7. FileVault on, firewall on with stealth, Handoff and Universal Clipboard off so the clean account can't see the dirty one's paste buffer. 8. Patch. OS, browser, router. Let it reboot.
137
INTEL: BTCPay Server hot wallets drained by attackers across multiple instances, users told to shut down servers
46
69
25
408
62,290
Sam Noble retweeted
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5: hackertube.org/ - Excellent recs without any logins (state is kept in local browser session) - Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads - Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version - HackerTube Longs ✊ A bit more on the process: I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing. Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
2,415
4,870
1,831
52,693
5,969,078
This is generally good advice, but you need to understand why. Tailscale’s WireGuard tunnels run over UDP, so this TCP bug has no impact. With no public TCP listeners open, there's nothing for the attack to hit. DERP fallback is outbound HTTPS to their relays, not an inbound TCP port on your box. Correction: Tailscale does not require an open inbound UDP port. It uses NAT traversal / hole punching. Outbound UDP from 41641 is sufficient in most cases; the stateful firewall allows the return traffic. Opening 41641 inbound is optional and only improves the chance of a direct connection instead of falling back to DERP. "Close all ports" is not a complete control and largely impractical for most public facing servers. Firewalls treat TCP and UDP separately. Most cloud security groups and host firewalls default to blocking inbound TCP while still allowing the UDP return path created by hole punching. If this bug had been in UDP, that distinction would have mattered. You're only right until you're not.
This is why I said you want your VPS ports all closed Then only let it be accessible inside your Tailscale for SSH and via a Cloudflare Tunnel to receive web traffic Any Linux box can be taken over with just a single TCP port exposed to the internet now A stop gap is disabling MPTCP on your VPS now but even better is not having any public ports exposed!
133
Sam Noble retweeted
🚨 PoC RELEASED: CVE-2026-112487 (macOS Storage) A path handling flaw lets an app already on a Mac gain root privileges. Public exploit code from a credited researcher appeared on 31 Oct. Fixed: Golden Gate 27.2, Tahoe 26.7.2, Sequoia 15.8.2 (26 Oct) 
Status: no NVD score yet, not in CISA KEV, no known in-the-wild attacks Action: update macOS now. threatwire.tech/news/update-… #macOS #CyberSecurity #poc
5
29
4
156
24,254
C'mon, at least we've progressed from 2014 tech consulting! Slow database? Migrate to cloud. Too many requests? Migrate to cloud. Tasks taking too long? Migrate to cloud. One server struggling? Migrate to cloud. Repeated failures? Migrate to cloud. Traffic getting expensive? Migrate to cloud. Data getting huge? Migrate to cloud. The early 2020s were wild. Anyone else have fun with: slimming down fat SQL queries; introducing connection pooling; optimizing flame graphs; unblocking CPU and or memory bottlenecks; adding transient error handling; and reducing payload size? Or was that just me?
System design is basically: Slow database? Add a cache. Too many requests? Add a load balancer. Tasks taking too long? Add a queue. One server struggling? Add more servers. Repeated failures? Add retries. Traffic getting expensive? Add a CDN. Data getting huge? Add database sharding.
1
3
381
I love how AI is lifting the curtain on the nonsense I’ve fought my whole career. All those meetings where I got dragged over the coals for “slow” response times because someone had to travel to a secure location or get in front of an authorised piece of hardware. Multi-layered processes called inefficient because they took too long. Security is insurance. As a CTO you get no praise when it works. You only hear about it when it doesn’t. Now the world is burning and everyone’s freaking out at the harsh reality. We didn’t build those layers for fun. We’d seen what lives under the fantasy of frictionless access. The problem isn’t that ~1,100 engineers can hit prod from a laptop. It’s that nobody told them they couldn’t. No process gets followed when there isn’t one. Vibe-coded apps are shipping GDPR breaches faster than anyone can count. Web3 is worse. On and off, I’ve spent the last 8 years in India. Nobody worries about traffic fines because there aren’t enough cops for the drivers. That’s tech right now.
58% of almost 2000 engineers can access database from their laptops, what's the impact: - companies do not care about data/security - one hacked laptop and customer data is leaked - gdpr fines can kill small company - one wrong query and prod data is gone It's not if, it's when.
1
1
108
Love this topic, I've spent a lot of time discussing this. It's not black and white... tl;dr: Trust but verify. The problem is not new. The trickle of code into production is now a flood. I have run teams up to a couple of hundred people. You do not (and cannot) read every line yourself. You build the system that does. Those systems have not caught up with the new volume. The same pattern is showing up in math, science and medicine. There has always been questionable code in production, the difference we're seeing now is volume and velocity. Investors have always wanted to ship as fast as possible and now the path of least resistance is wider. What used to be a stream of hand-written code is a flash flood of LLM-generated code. More questionable code landing in prod means more code for attackers to analyse, and they can do that faster too. The models are not magically inventing worse code. They were trained on human-written code. I have led teams across the globe for years and have seen first-hand the volume of low-quality output that almost certainly makes up a big chunk of that training data. Now the rest of the world is generating (and shipping) that style of code at scale. Same coin, different sides. Also note that plenty of the big exploits we have seen recently sat dormant for years, long before “vibe coding” was a term. Dirty COW (CVE-2016-5195, in the kernel since 2007). Dirty Pipe (CVE-2022-0847). The 15-year-old iSCSI issues (CVE-2021-27365 et al.). Linux kernel bugs hide about two years on average. The low-level stuff that runs the world has always been a mess. We just did not know the full extent. What LLMs change is that verification has not scaled with the output. Anyone can now vibe-code something and ship it. The models do what you ask and rarely volunteer the security questions or the cross-checks unless you action them. Sensitive paths still need multiple review rounds across different models built around battle tested processes. Trust but verify, the same way you run a multi-disciplinary engineering team. The CTO of a suitably large enough organisation doesn't read every line of code. They build the system that does. This is not unique to tech. The same pattern is playing out in math and medicine: LLMs accelerating real advances, same tools, same incentives, same gaps in verification, different domain. The code quality question is still open. The volume + missing process problem is not. This is more like cars getting fast enough that seatbelts stopped being optional. The kinetic energy was always there. The speed made the safety system mandatory.
I'm relatively confident in predicting that currently, more major breaches are likely to be the result of AI use by the target, not the attacker. Ransomware actors have been consistently winning for over a decade. They've had so much time to perfect their tradecraft, and non-AI ransomware actors can now blast through entire networks in minutes. It is extremely unlikely that organizational security has, in 4 years, after decades of failure, matured enough that hackers now somehow need AI to win. What has likely happened, though, is that organizations have spent the last 4 years falling over themselves to shove AI into everything, negating longstanding security fundamentals. "only give a user/system the minimum possible access needed to do its job" has become "just give the AI access to everything and give everyone access to the AI". "use processes with well-defined and easily auditable input and output" has become "the AI correctly processes requests all the time, 60 percent of the time" "reduce attack surfaces" is now "every department should be vibe coding. It doesn't matter if you've never coded before and don't understand information security. Just make public-facing dashboards, it'll be fine." I'd say if anything, hackers aren't getting measurably more sophisticated, but the average organization's security posture has gotten exponentially worse.
4
1
7
1,351
Awesome detailed breakdown of the real risk for anyone using an older iPhone as a cold / warm wallet. tl;dr; Older iPhones (roughly 4S through the 11 series) have unpatchable BootROM flaws (Apple cannot fix these). The exploit needs physical access and is far less dangerous if you use a strong wallet password + long alphanumeric passcode instead of a 6-digit PIN. Security is layers, max out every one of them and keep your devices updated (both the OS and installed software). Beyond the tl;dr: once someone has the phone in hand they can put it in DFU and run code before iOS loads. Passcode and Secure Enclave remain in play, but a weak passcode (4 or 6 digits) isn't a real barrier. I always treat any phone that has held a seed as potentially extractable the moment it leaves controlled hands. New seed generated clean, long passcode, only ever use your own cables, and migrate if it’s ever been out of sight. If you have a pre-iPhone 12 device you’re using as a cold wallet and aren’t sure of the above, move the funds to a new setup.
Over the past two weeks iPhones have been publicly jailbroken more than once, and most people treated it as a bit of tech drama and scrolled on. I want to use it to talk about something that matters more to people in crypto: wiping an old iPhone, putting it in airplane mode and using it only to hold your seed phrase is not as safe as most people think. If it's an iPhone 11 or older, the problem lives in the chip, and no iOS update will fix it. Quick recap of what happened. On October 16 at Hexacon in Paris, the Calif team jailbroke an iPhone 18 Pro running iOS 27.0.1 live on stage. The jailbreak survived a reboot, they had root, and they got past MIE, the hardware memory protection Apple added starting with the A19 specifically to stop spyware. In the same session they showed attacks that get in through a single click in Signal or iMessage. A week later, on the day iPhone Duo went on sale, they jailbroke that too. Apple shipped iOS 27.2 on October 26, and Calif says most of the bugs they used are now fixed. Then on October 30, Ikotas Labs, the Japanese team that just won Pwn2Own, posted that they had jailbroken 27.2 as well. Their video only shows swapped icons and the version number, though, so I'm treating it as unconfirmed for now. For most people on a recent phone who keep it updated, none of this is a reason to panic. Before chains like these get patched, they usually sit with spyware vendors and forensics companies, and the odds of an ordinary person being specifically targeted are low. Update to 27.2 and you've done what you need to. The people I actually want to warn are those using an old iPhone as a cold wallet. iPhone 4S through iPhone X have checkm8 in the BootROM, made public in 2019. This June, usbliter8 went public, covering the iPhone XS, XR, the 11 series and the 2nd-gen SE. The BootROM is boot code burned into the chip at the factory. Apple can't patch it with a software update, so these devices will carry the hole for as long as they exist. Exploiting it doesn't need a network either. Once someone has your phone in hand, they put it into DFU mode and plug in a custom USB board (the public research used a $5 Raspberry Pi Pico 2), and they can run their own code before iOS even boots. Airplane mode makes no difference here. I don't want to overstate it. On models like the XS and 11, the BootROM bug alone won't hand over your seed phrase, because your passcode and the Secure Enclave are still in the way. The trouble is that these remaining layers are exactly where people tend to cut corners: a 6-digit passcode, an iOS version years out of date with a pile of public n-days, a short wallet app password. On some older models, forensic tools use bugs like this to brute-force the passcode. And if your wallet app encrypts the seed with nothing but a short password, once that encrypted file is copied off, the attacker can take it home and keep guessing offline, with no attempt limit. Old iOS versions also have a charging port problem. ChoiceJacking, published in 2025, lets a malicious charger pose as a keyboard and approve the data-access prompt for you. Apple only closed that in iOS 18.4. One more thing about this kind of attack: when you get your phone back, nothing looks off. In security circles it's called an evil maid attack. A hotel room, a security check, lending your phone to someone to charge — any time it's out of your hands for a while, there's an opening. A lot of people only find out their phone was touched after their assets are already gone. My advice: 1. Check which iPhone holds your seed phrase (Settings → General → About). If it's an iPhone 11 or older, move to a newer model before storing a seed on it. 2. If you've already stored one there, no need to panic. When you get a chance, generate a brand-new seed phrase in an environment you trust and move your assets over. It has to be a whole new seed. A fresh address under the old seed won't help. 3. If your phone has ever been out of your hands for a while, say held for inspection at a border, don't wait. Migrate as in step 2 as soon as you can. 4. Update your daily iPhone to iOS 27.2 (Settings → General → Software Update). If you haven't moved to 27 yet, install at least 26.7.2. 5. Switch to a long alphanumeric passcode (Settings → Face ID & Passcode → Change Passcode → Passcode Options). A 6-digit PIN has only a million possible combinations. Make your wallet app password longer too. 6. Carry your own charger. Don't use public USB ports, random power banks or cables you don't know. 7. If you hold a lot or are likely to be targeted, consider turning on Lockdown Mode (Settings → Privacy & Security → Lockdown Mode). It blocks wired connections while the phone is locked. 8. Any "iOS 27 jailbreak download," or any tutorial asking you to install a certificate or configuration profile, is a scam or malware. Neither Calif nor Ikotas has released any tools. If you like using an iPhone as a cold wallet, go with a newer model and keep it on the latest iOS. There's no public BootROM exploit for the iPhone 12 or later, and if something does turn up, Apple can still fix it with an update. With really old devices, the risk sits in the chip itself, and updating iOS won't get you around it.
3
3
311
This is pretty nasty. If you're using Chrome (or a Chromium based browser a la Brave) make sure you're up to date. For Chrome that means >= 155.0.8059.39 For Brave that means >= v1.97.56
Introducing CVE-2026-106378: Chrome GPU Sandbox Escape to full Remote Command Execution using synthetic keyboard injection By combining it with common GPU bugs, it becomes a full chain. Binance DevOps was most likely pwnd with a zeroday chain affecting the browser 😬
2
130