Block Max

Ces derniers temps, je passe beaucoup de temps à faire défiler les nouveaux projets L2 et rollup, et un sujet revient sans cesse dans mon fil d’actualité que je n’arrive pas vraiment à ignorer — Newton Protocol. Si vous ne l’avez pas encore croisé, il s’agit essentiellement d’essayer de construire l’infrastructure permettant d’envisager un monde où des agents d’IA gèrent réellement l’argent onchain, et pas seulement d’en bavarder. Et plus j’ai creusé, plus je me suis rendu compte que ce n’est pas un vague combo “IA + crypto” sorti de nulle part. Il y a une vraie architecture derrière.
Permettez-moi de faire un pas en arrière. On a tous vu une centaine de projets coller “IA” sur leur livre blanc cette année. La plupart se résument à une enveloppe pour chatbot ou à un générateur de signaux de trading que personne ne peut vérifier. Ce qui m’a interpellé avec Newton, c’est que ce n’est pas l’IA qu’ils vendent comme produit. Ce qu’ils vendent, c’est la vérification en tant que produit, avec des agents d’IA comme l’un des cas d’usage qui repose dessus.

Crypto solved the easy problem first. That's the thing I keep coming back to. We made execution trustless. Anyone can send anyone else money without a bank standing in the middle, without some institution deciding your transaction doesn't fit their risk model today. Fine. Good. That got solved years ago. But the part that comes after — who, or what, gets to act on your behalf, and how do you actually know it's doing what you told it, not what it decided to do on its own — nobody really cracked that one. That's the gap Newton Protocol is trying to climb into. And look, the branding around it doesn't do it any favors. AI agents, rollups, verifiable this, verifiable that. It reads like buzzword soup. But underneath the soup, there's a real problem here, and I think it's more underrated than people give it credit for.
Here's what actually got my attention. Newton isn't a trading bot platform. People keep calling it one, and sure, a lot of the use cases are trading-adjacent — recurring buys, rebalancing, yield chasing, copy trading, orders that fire when a price crosses some line. But that's not the core of it. The core is a permissions layer. A way to tell an AI agent "do this, but only under these conditions, and don't you dare touch anything outside this box" — and then actually prove that promise held, instead of just hoping it did. Trusted execution environments handle the offchain computation. Zero-knowledge proofs verify it followed the rules. That's the skeleton. I won't pretend TEEs make for thrilling reading. They don't. But the instinct behind it is solid: if you're handing software access to your money, a nice dashboard and a promise shouldn't be enough anymore.
The team behind this is Magic Labs, a company that's been building wallet infrastructure for a while now. Alongside it sits a separate nonprofit, the Magic Newton Foundation, meant to carry the community and decentralization side of things. That's a familiar split in this industry — company keeps building, foundation handles the "we're decentralizing, promise" narrative. Whether that split actually plays out the way it's pitched, I don't know. Nobody does. Roadmaps are wish lists with dates attached, and crypto has a long, ugly history of "progressively decentralized" quietly becoming "still centralized, three years later, with a governance forum nobody bothers posting in." That's not a shot at Newton specifically. It's just what the word "progressively" tends to mean once the marketing wears off.
So what's actually under the hood? Three main pieces, more or less, though the exact framing shifts depending on which writeup you're reading and when it was published — which is its own small tell about how fast this thing is still moving. First, the Model Registry. An onchain catalog where developers publish "agent models" — smart contracts that encode trigger-action logic. Price drops ten percent, execute a trade. Volatility spikes, rebalance. RSI dips below some level, buy. Simple enough on paper. The real complexity shows up once developers start stacking these models on top of each other in ways nobody planned for. That's always where the real test happens. Not the demo. The moment some user chains four agents together and something behaves in a way none of the original builders saw coming.
Second, the Keystore — a dedicated rollup that handles permissions. Who's allowed to do what, under what conditions, how those rules get issued or pulled back. Some material calls this zkPermissions specifically, letting users set granular limits — spending caps, timing windows, conditional triggers — enforced by math instead of a terms-of-service page nobody reads. Third, the account layer, built on ERC-4337 smart accounts, which lets you delegate a narrow slice of control instead of handing over your whole wallet and crossing your fingers. This last piece doesn't get talked about enough, honestly. The difference between "here's my private key, good luck" and "here's a scoped, revocable permission that can't touch anything outside its lane" isn't small. It's the entire difference between something you'd trust with real money and something you'd only ever test with money you can afford to lose.
There's also a newer angle showing up more in their messaging lately: compliance as code. Builders write policies — apparently in something like Rego, the same language a lot of cloud infrastructure already leans on — and a decentralized network of operators checks every transaction against those policies inside secure enclaves, spitting out a cryptographic proof that the check actually happened, verifiable by anyone through their explorer. My first reaction to seeing "compliance" as a headline feature of a crypto protocol was a wince. Compliance is usually the word that shows up right before decentralization quietly slips out the back door. But sitting with it longer, I'm not sure that reaction holds up here. If the real goal is getting actual financial institutions, stablecoin issuers, real-world-asset platforms comfortable enough to let AI agents touch real money — and that does seem to be who they're courting now — then some kind of programmable, verifiable compliance layer isn't optional. It's closer to a prerequisite. Whether it stays neutral the way they're describing it, or slowly turns into a chokepoint controlled by whoever writes the dominant policies — that's the real question. Nobody can answer it honestly yet. Ask again in two years.
The token, NEWT, does what you'd expect. It's gas for the rollup — issuing, updating, revoking permissions and session keys all cost NEWT. It's collateral for operators running agents, with slashing built in if an agent misbehaves, which at least makes running a sloppy or malicious agent cost something real instead of nothing. It secures the Keystore rollup through delegated proof-of-stake, with a fourteen-day unstaking window that's pretty standard for this kind of design. And eventually it carries governance weight, once that side actually turns on. There's a fee market modeled loosely on Ethereum's EIP-1559, meant to keep ordering fair and stop the network from clogging during busy stretches.
Total supply is capped at one billion tokens. No inflationary minting, which removes one long-term worry even if it does nothing for the short-term ones. The split leans community-heavy on paper — around sixty percent earmarked over time for rewards, staking incentives, and grants, rather than sitting entirely with the founding team. At launch, only a modest slice — somewhere in the low twenty percent range — actually entered circulation. The rest sits behind a fairly standard vesting structure. Team and early backers both behind roughly a twelve-month cliff, then unlocking linearly over another three years or so. And here's the make-or-break moment nobody likes talking about out loud: a big locked supply plus a public unlock calendar is basically a countdown clock the whole market watches. That's not a knock on Newton specifically. It's just what happens to any token built this way. There was already a sizable unlock flagged for late January 2026, and events like that tend to bring sell pressure whether or not real usage has grown enough to absorb it.
On the trading side, things moved faster than the actual protocol did — and that's a pattern I notice constantly in this industry, and it never stops bugging me a little. NEWT landed derivatives listings on major exchanges pretty quickly after launch, including a leveraged perpetual contract on Binance, funding settled every few hours, all the usual mechanics of a modern perp market. That brings liquidity, sure, and attention. But it also means a huge chunk of the price action early on has basically nothing to do with whether the Model Registry is actually composable or whether the Keystore is decentralizing on schedule. It's leveraged traders reacting to leveraged traders. A separate ecosystem that just happens to share a ticker with the infrastructure underneath it. Worth remembering next time the price does something wild and someone tries to read deep meaning into a candle.
Regulatory positioning is one of the more interesting threads here, mostly because most projects at this stage haven't bothered thinking about it publicly at all. Newton's apparently already secured some form of classification under the EU's MiCA framework — filed under something like "other crypto-asset." Whether that ages well as enforcement matures, or as U.S. regulators finally land somewhere solid on autonomous financial tools, which has been a painfully slow and genuinely unresolved conversation — that's anyone's guess. But engaging with it early instead of waiting for a subpoena to force the issue says something. It reads like a project aiming at institutional plumbing more than retail speculation, even while retail speculation is clearly what's driving the token's day-to-day price right now.
Here's where I land, honestly uncertain, and I don't think that's a bad place to be. It's not the cryptography or the architecture that worries me — that's been through independent audits, reports are public, and on paper it holds up, or at least holds up as well as any young, still-shifting codebase can. What I'm actually unsure about is adoption. The boring kind. Does a developer really choose to build an agent inside this Model Registry, instead of just writing their own bot and running it off a rented server somewhere, quiet, off to the side, answerable to nobody — the way basically every trading bot has ever been built? The pitch is that verifiability and composability make this categorically better, not just marginally better. Maybe that's true. But "categorically better" claims about crypto infrastructure have a rough track record of turning into developers actually showing up. Plenty of technically superior systems shipped into total silence because switching cost too much friction, or there was no existing user base to plug into, or the old centralized way was just good enough for most people's purposes. That kills momentum before it ever really starts.
Then there's the marketplace — still "upcoming," not live, as of the latest updates I can find. The public place where you'd go discover an agent someone else built, activate it, maybe pay for it, maybe stack it with agents you're already running. This is the piece that, if it actually works, turns Newton from "interesting infrastructure a handful of protocols quietly plug into" into something with real network effects — the kind where every new agent published makes the whole thing marginally more useful for the next person walking in. But marketplaces are brutal. Genuinely, categorically harder than pure infrastructure. Getting supply and demand to show up at the same time, in the right proportions, with enough quality on the supply side that early users don't bounce after one bad experience and never come back — that's a cold-start problem that's quietly killed way more ambitious platforms than this one. In crypto and well beyond it.
So no, I'm not handing you a clean verdict here. And honestly, I'd be a little suspicious of anyone who does hand you one this early, about a protocol whose most consequential pieces — the fully multichain rollup, the open marketplace, the shift to a permissionless validator set — are still sitting in the "upcoming" column of their own roadmap. But here's what I do believe, with more confidence than the rest of this: the problem Newton keeps circling is real. Maybe more real than most of what gets funded and hyped in this corner of the industry. Automation without verification is just trust wearing a costume made of code. And trust wearing a costume is exactly the thing that's burned people over and over — every time some offchain bot got quietly compromised, every time some centralized automation service rugged its users while the landing page still said "secure." If Newton, or something built roughly like it, actually manages to make "the agent did exactly what it was permitted to do, and here's the proof" a normal, boring, expected feature of onchain finance instead of a headline on a pitch deck — that would matter. Not because it's flashy. It wouldn't be flashy at all if it worked the way it's supposed to. It'd be the kind of thing nobody notices when it's working and everybody notices the second it isn't. That's usually the real tell, in my experience, in any corner of technology you look at closely enough. Whether this particular team, this particular token, on this particular timeline actually gets there — or gets there first — is a whole separate question. And honestly, I'd rather watch that play out than pretend I can call it right now.
@NewtonProtocol #Newt #newt $NEWT

"Automatisation sans confiance." J’ai tellement lu cette phrase qu’à force, elle finit par presque ne plus me toucher. Mais prenez une seconde pour y réfléchir : elle veut dire quelque chose. Le protocole Newton fait partie de ces projets qui vous obligent à ralentir et à vous demander : quel problème est-ce qu’on essaie vraiment de résoudre ici ? Car, sur le papier, on dirait n’importe quel autre pitch « IA + crypto » des deux dernières années. Des agents. Des rollups. Des places de marché. Les mots se mélangent. Je dis ça en tant que personne qui a lu beaucoup de ces présentations, et je dois admettre qu’une certaine fatigue s’installe à partir du cinquième slogan « onchain alimenté par l’IA » de la semaine.

La crypto a passé une décennie à résoudre le problème de la monnaie sans confiance. Cette partie est essentiellement faite. Maintenant, elle bloque sur quelque chose de plus difficile : l’action sans confiance. Envoyer un jeton d’un portefeuille à un autre ? C’est facile. Il a fallu des années pour le construire, certes, mais le concept est simple. Laisser quelque chose d’autre agir à ta place — un script, un modèle, un agent, peu importe comment tu veux l’appeler — c’est une autre affaire. Tu confies le contrôle pendant que tu dors. Et il faut que ce soit vraiment sûr, pas seulement probablement sûr. C’est l’écart que le protocole Newton essaie de combler. Avant de m’enthousiasmer ou d’en douter, je veux prendre le temps de comprendre pourquoi cet écart est si difficile à combler en premier lieu.

There's a specific kind of discomfort that comes from handing part of your financial life to something that isn't a person. Crypto has never really solved that. Not once. Bots have been running this show since the early days — arbitrage scripts, liquidation keepers, market makers humming along in the background of every exchange you've ever touched. But they operated in a black box. The "trust" you gave them was really just trust in whoever wrote the code and held the keys. That's it. That's all it ever was.
Newton Protocol shows up and makes a claim that sounds almost too neat: what if the leash itself was provable? Not "trust the team." Not "we got audited, we're fine." A live, checkable, cryptographic answer to one question — did this thing only do what it was allowed to do. That's the whole pitch. Everything else is scaffolding built to make that pitch real instead of just marketing copy.
Let's get into the mechanics, because the language around "AI plus blockchain" has gotten so mushy that words stop meaning anything. At the core is the Newton Keystore, a rollup — an L2, in the usual sense of batching computation off-chain and settling proofs back onto it. But here's the thing: most rollups try to be everything to everyone, general-purpose execution machines. This one doesn't. It has one job. Permissions. It's where the rules governing an AI agent, or a delegated bot, or a smart account actually live, and where they get updated when something changes. Pair that with ERC-4337 smart accounts — the standard that splits wallet authority into fine-grained pieces instead of an all-or-nothing key — and you get what they call zkPermissions. Rules like "only trade if volatility crosses X" or "cap daily spend at Y," enforced not by some script running on a guy's laptop but by a system that spits out a proof every single time it acts.
Now, the engineering underneath that layer is where things get more debatable. Newton leans on trusted execution environments — sealed hardware enclaves where computation happens out of public view — paired with zero-knowledge proofs that confirm, after the fact, the enclave did what it was supposed to. This isn't unique to Newton. It's becoming the standard playbook across the industry. But I want to be straight about what you're actually trading here. A TEE is a hardware trust assumption. You're not trusting a person anymore, or a company. You're trusting that a chip manufacturer's secure enclave hasn't been compromised, doesn't have some undisclosed flaw, hasn't been quietly backdoored by someone with the access and the motive. That's a different kind of faith than reading open-source code yourself. Smaller than the blind trust bots have always demanded, sure. But not zero.
My honest take? This is a bridge, not a destination. Fully verifiable machine learning, done entirely in zero-knowledge with no hardware assumption at all, is still too slow and too expensive for real-time trading decisions. TEEs get you something that actually works today. Whether Newton eventually walks away from that hardware dependency as zkML matures — which a lot of people in this space expect the whole category to do eventually — is one of the bigger open questions hanging over the long-term design. Nobody's answered it yet. Maybe nobody can, yet.
Then there's the Model Registry, and this is the piece aimed at builders rather than end users. Honestly? I think this is the part that decides whether Newton becomes real infrastructure or just another protocol nobody touches after the airdrop farmers leave. It's an on-chain marketplace where developers publish "agent models" — reusable chunks of trigger-and-action logic, basically smart contracts encoding a strategy, that anyone can activate or build on top of. Recurring buys with cryptographic proof of exact timing and price. Yield agents shifting capital across protocols as rates move. Vault managers watching collateral ratios, stepping in before a liquidation hits. Copy-trading setups that mirror another wallet while still respecting your own limits.
On paper, that's genuinely interesting. It's the difference between a protocol that just processes transactions and one where value actually compounds — developers building on each other's work instead of everyone reinventing the same rebalancing bot for the hundredth time. But look, marketplaces are brutal to bootstrap. You need enough builders publishing strategies worth using to pull in real users. And you need enough real users for publishing to be worth a developer's time in the first place. That chicken-and-egg problem has quietly killed more ambitious crypto marketplaces than any smart contract bug ever has. Does Newton's registry clear that bar? Or does it end up another mostly-empty directory with a few demo strategies gathering dust? I don't know. And I'd be suspicious of anyone who tells you they do, including, probably, the team building it.
Here's the tension I keep coming back to, and it's not just a Newton problem — it applies to everything chasing this same idea. Verifiability tells you an agent followed its rules. It says nothing about whether those rules were any good. If I write a permission that's too loose — a stop-loss with a blind spot, a volatility trigger that doesn't account for some specific kind of shock — the system will faithfully, provably execute exactly what I told it to do. And then hand me a cryptographic receipt proving it lost my money exactly the way I authorized. That's not a bug. It's baked into the entire concept of programmable automation. You can trace this straight back through every DeFi exploit that technically wasn't a hack at all — just code doing precisely what it was written to do, in a situation nobody thought to model. Newton protects you from a rogue agent going off-script. It cannot protect you from yourself writing a bad script in the first place. That distinction matters more than most of the "trustless automation" marketing language wants to admit.
On the token side: NEWT has a fixed supply, one billion total, with roughly sixty percent carved out for community stuff — staking rewards, grants — and the rest split between core contributors and early backers on multi-year vesting. At launch, only a small slice was actually circulating. The rest sits behind cliffs and linear unlocks stretched over three to four years, depending on the bucket. To its credit, the token isn't just a speculative chip sitting there for no reason. It does real work. It's gas for issuing and updating permissions. It's staked by validators securing the Keystore under a delegated proof-of-stake model with a two-week unstaking window. It's collateral posted by operators running agents, subject to slashing if those agents misbehave. And eventually, it's the tool for governance once that layer actually goes live.
None of that changes the calendar problem, though. A big chunk of total supply is still working its way out of lockup, and unlock schedules in this industry have a well-worn habit of triggering sell pressure that has nothing to do with how the product is actually doing. I've watched this movie enough times to be wary of reading too much into short-term price swings in either direction — bullish or bearish — while most of the float hasn't even hit the market yet.
The part of Newton's current pitch that I think is genuinely underrated — especially next to how much air the "AI agents" framing sucks up — is the compliance angle. The team's more recent messaging leans hard into compliance-as-code: builders write policies in something like Rego, a decentralized set of operators evaluate them inside that same TEE-secured environment, and anyone can verify the results through a public explorer instead of just trusting the process. Strip away the branding, and this is a real gap in the market. Institutions, stablecoin issuers, platforms dealing in tokenized real-world assets — they don't avoid on-chain automation because they hate efficiency. They avoid it because a compliance officer can't sign off on a system where enforcement is basically vibes and a Discord mod keeping an eye on things.
The real clincher here is this: if a protocol can turn "did this transaction meet our regulatory obligations" into something as checkable as "did this transaction settle" — the same way smart contracts made execution itself checkable over a decade ago — that's a meaningful unlock for the kind of capital that's mostly sat on the sidelines of DeFi. I wouldn't call this winner-take-all. And I wouldn't bet confidently that Newton specifically becomes the dominant neutral layer here, rather than one of several standards that end up coexisting. That's still being fought out through shipped code and actual adoption. Not whitepapers. Not announcements.
So where does that leave me? Somewhere between genuinely curious and deliberately unconvinced — which, honestly, is the only sane place to stand with infrastructure this young. The problem being solved is real. Offchain, unverifiable automation is a live liability across this whole industry, not something dreamed up to justify a token launch. The technical approach is reasonable, not just acronym soup dressed up to look smart, even if leaning on trusted hardware right now is a compromise I'd expect to soften as zero-knowledge machine learning gets cheaper and faster.
But the pieces of the roadmap that actually matter — a marketplace with real usage, a fully multichain permissions rollup, a validator set that's decentralized in practice and not just on paper — are still sitting in the "upcoming" column. Which means a good chunk of the current sentiment around this thing is a bet on future execution, not a reflection of activity that already exists. That doesn't make it a dead end. It doesn't make it a sure thing either. It makes it early, in that specific way infrastructure always looks early right up until the moment it suddenly isn't — and there's rarely a clean signal telling you which side of that line you're standing on while you're still standing on it.
This isn't financial advice, and given how young and thinly traded this token still is, I'd raise an eyebrow at anyone writing about it who claims otherwise. What's actually worth watching is the gap between what Newton already does and what it still just says it will do. Because that gap, like it usually is with projects like this, is where the real risk is quietly sitting. Waiting.
@NewtonProtocol #Newt #newt $NEWT

Il y a une phrase que l’équipe Newton répète sans cesse. La crypto a construit la maison de verre. Newton construit les verrous. La première fois que je l’ai lue, j’ai levé les yeux au ciel. Ça ressemblait à quelque chose qu’aurait concocté une personne du marketing un mardi matin lent. Mais elle m’est restée. Et plus je la mâchais, plus je me rendais compte que ce n’est pas juste une formule astucieuse. C’est en fait tout le problème, énoncé d’un seul souffle.
Voici ce que ça veut dire. Tout, dans la crypto, est visible. Chaque portefeuille, chaque transaction, chaque bêtise, tout est là, au grand jour, pour toujours. Une transparence totale. Et pourtant, rien n’a jamais été sûr. Tu pouvais regarder le vol se produire en temps réel et, malgré ça, ne pas réussir à l’empêcher. C’est le paradoxe. Des murs de verre, sans serrures. Newton essaie de construire les verrous. Qu’ils y parviennent ou non, c’est une autre question. Mais c’est la bonne question à se poser.

Il est étrange de regarder en arrière vers les débuts de la crypto et de se rendre compte à quel point il fallait réellement de travail manuel dans ce que nous appelions « l’automatisation ». On restait là, à fixer des graphiques, à exécuter des transactions à la main, en priant pour que la connexion Internet ne se coupe pas au moment même où le marché allait contre nous. Et maintenant, nous nous tenons au bord de quelque chose d’entièrement différent avec le Newton Protocol, ou NEWT, comme l’indique le ticker. Un projet qui tente de résoudre l’un des problèmes les plus chaotiques à l’intersection entre l’intelligence artificielle et la finance décentralisée. À mon avis, l’industrie danse autour de ce sujet depuis des années, cherchant à greffer l’IA sur la blockchain comme on monterait un silencieux de rechange sur une voiture, mais ça ne s’emboîte jamais parfaitement, parce que l’infrastructure de base n’a pas été conçue pour la manière dont les agents d’IA doivent fonctionner : rapidement, de façon autonome et avec une logique implacable. Alors que la blockchain est souvent lente, congestionnée et coûteuse à l’excès dès qu’on tente d’y faire passer une logique complexe. Newton vise à être un rollup sécurisé, spécifiquement pour ces stratégies pilotées par l’IA et le trading automatisé. Et honnêtement, il était temps que quelqu’un s’attaque à l’éléphant de la sécurité dans la pièce : laisser un bot d’IA contrôler vos clés privées sur une chaîne standard, c’est une perspective terrifiante pour quiconque a au moins la moitié d’un cerveau et un portefeuille qui lui tient à cœur.

Je fais défiler mon fil d’actualité depuis quelques mois, regardant chaque projet se rebaptiser « propulsé par l’IA », exactement comme tout le monde a rebaptisé « DeFi » en 2020 et « metaverse » en 2021. La plupart de tout ça, c’est du bruit : un chatbot greffé sur un DEX, un bot de trading avec un compte Twitter, un livre blanc qui dit « agent » quarante fois. Donc, quand je suis tombé pour la première fois sur Newton Protocol (NEWT), ma réaction honnête a été le scepticisme. Encore un rollup d’IA. Super. Mais plus je me suis plongé dans ce qu’ils essaient vraiment de résoudre, plus je me suis surpris à hocher la tête au lieu de faire défiler.

Je me pose cette question depuis un moment maintenant, une question qui revient sans cesse dès que j’essaie vraiment d’utiliser la DeFi comme elle est censée être utilisée. Pourquoi gérer un portefeuille solide sur plusieurs chaînes donne encore l’impression d’un travail à temps partiel ? Faire des ponts ici, réclamer des récompenses là-bas, rééquilibrer ailleurs, et regarder les frais de gaz grignoter la moindre partie du rendement que vous poursuiviez à la base. Si vous avez déjà essayé d’exécuter une stratégie ne serait-ce que légèrement sophistiquée sur trois ou quatre protocoles, vous voyez exactement ce que je veux dire. Alors, quand je suis tombé sur le Newton Protocol et son jeton NEWT, ma première réaction a été le scepticisme. Encore un projet d’"automatisation", je me suis dit. On a déjà vu des bots. Mais plus j’ai creusé, plus j’ai compris que celui-ci tente de résoudre un problème qui m’embête depuis longtemps — pas seulement de l’automatisation, mais une automatisation en laquelle vous pouvez réellement avoir confiance.