aijia

The DeFi community talks too much about 'audits' and 'insurance'—in essence, they are all post-incident remediation, following the same logic as auto insurance. But the Newton Protocol Mainnet Beta takes a different approach: it blocks risks before transactions are even put on-chain. It positions itself as an 'onchain authorization layer for DeFi'—not something that lets you collect evidence to file a claim after things go wrong, but instead runs a complete strategy check before every transaction is executed. Are there issues with KYC/KYT? Does the wallet risk score pass? Is the credit rating sufficient? Has the price feed deviated? Is the target contract flagged as high-risk? Only after these checks pass will the transaction be signed and allowed to proceed. For institutional funds, this is crucial. Traditional DeFi treasuries mostly rely on manual approvals—low efficiency, and someone inevitably ends up taking the blame. Newton's VaultKit turns this entire workflow into composable strategy modules. Chainalysis can be plugged in for contract risk, RedStone for price feeds, Credora for credit, and Webacy for wallet risk. These proven components can be used directly. Most importantly, every approval decision leaves an on-chain signature record—auditable, accountable, and able to prove innocence. Against the backdrop of tightening regulation, institutional funds want to go on-chain but don’t dare—what they lack is exactly this kind of infrastructure for proactive defense. A project incubated by Magic Labs, with participation from PayPal Ventures—the experience from 57 million wallets wasn’t gained for nothing. $NEWT #Newt @NewtonProtocol

Everyone who does DeFi knows that no matter how much TVL gets poured in, if risk control can’t keep up, problems will show up sooner or later. @NewtonProtocol's VaultKit is specifically built for this—helping vault managers enforce compliance and risk-control rules on-chain without having to write code from scratch. Chainalysis provides smart contract risk monitoring, RedStone supplies price feeds, Credora manages credit ratings, and Webacy checks wallet risk scores—these are strategy modules you can plug in directly within VaultKit. The key point is that what it records isn’t just “something went wrong,” but rather “which strategy rejected each transaction”—and that’s exactly the kind of reassurance institutional investors need. And what Magic Labs has done speaks for itself too: 57 million wallets, and PayPal has invested in it—so they’re absolutely capable of building this. #Newt $NEWT

On-chain DeFi doesn’t lack TVL right now—it lacks safety controls before transactions occur. @NewtonProtocol’s mainnet Beta launched on June 23. What it does is completely different from traditional DeFi security tools: it doesn’t tell you after the fact what was wrong with the last transaction; instead, it checks strategies before each transaction is settled. Compliance, risk control, identity verification, and risk scoring are all run before execution—only then is the transaction allowed to proceed. This effectively automates an institution’s compliance review process and embeds it directly into the blockchain. Moreover, every approval decision has an on-chain signature record, so institutions can tell investors, “Look—every transaction has passed risk control,” without having to hand over a hundred-page compliance report. #Newt $NEWT

$XLM today directly pulled up 10%, with a trading volume of 470 million. Stellar—this long-established payments chain—has suddenly been targeted by capital. It might be that the cross-border payments narrative is back in play again. $PYTH followed up and rose 8.6%, with a volume of 90 million. The oracle/“oracle machine” sector has clearly seen capital returning recently; demand for data is growing more and more for Pyth in the Solana ecosystem. $SOL is holding around 75—up less than 2% but supported by a massive $3 billion in trading volume; on-chain activity hasn’t dropped. On the decliners list, TIA got hammered again, down 5%. Celestia has been falling since yesterday—the pace at which the modular narrative is fading is even faster than when it was pumping. VELVET has been down for two consecutive days. The former “limit-up妖王” (king of consecutive rallies) has now turned into the biggest decliner—everyone who chased the price got buried.

People are still discussing AI Agents at the stage of human-use Agents, but the real breakthrough is when Agents call each other. Your trading Agent needs to call an analysis Agent’s model to make signal judgments; the analysis Agent then needs to call a data Agent to fetch data and pull it on-chain. Every step costs money, and every step verifies that the result hasn’t been tampered with. This is @OpenGradient’s killer scenario: x402 charges per use to handle micro-payments between Agents; on-chain inference proofs ensure that the call results are trustworthy; and PIPE guarantees that intermediate data isn’t intercepted. Stacking these three protocols together forms the settlement layer of the Agent economy. Traditional AWS can’t do this—centralized cloud has no on-chain verification. When you call an API, you basically don’t know whether the other side is returning a real model output or cached old results. Web3 naturally solves $OPG #OPG

There are more and more AI KOLs and AI trading call bots out there now. The truth and the nonsense are all mixed together, and you can’t really tell the difference. What I hate most is posts like “AI says a certain coin will go up,” but it can’t provide any basis at all. The advantage of OpenGradient Chat is that its reasoning results come with on-chain proof—your can verify everything: the prompt, the model version, and the output. At least before I get scammed, I can check whether this AI is actually reliable. @OpenGradient $OPG #OPG

After the EU AI Act took effect last year, many companies using AI for credit approvals and insurance pricing started to struggle with a problem: regulators require you to explain how the AI makes decisions, but black-box models can’t explain themselves. @OpenGradient’s on-chain reasoning proofs here are not just technical showmanship—they’re a compliance tool. For every AI inference, the inputs, model version, and intermediate calculation steps all generate on-chain proofs. When regulators come to check, you can pull up the proofs directly—doing that is more effective than writing a hundred pages of compliance reports. MiCA also has similar requirements for AI applications in the crypto-asset space. Traditional AI firms would have to spend a fortune to plug this gap by building audit systems, whereas Web3 is naturally set up for it. This narrative is actually the angle OG finds it easiest for institutions to buy into $OPG #OPG

$BTC has fallen back to $59,000 again. It’s down nearly one point on the day. The $60,000 level has been grind-testing for half a month, unable to hold. ETH has also pulled back to 1590. Today’s real interesting part is the top gainer list: $LIT . Lighter, this DeFi protocol, is up 12%, with a trading volume of 50 million—arguably the only “real deal” altcoin with something to talk about today. Over on the privacy coin side, $ZEC has quietly gained 4%—Zcash is at 398. Recently, there are signs that funds are returning to the privacy narrative. On the losers list, VELVET has crashed, down 11%. The “妖王” that kept rallying a few days ago is now getting hit the hardest—classic pump-and-dump behavior. If a coin jumps big one day and dumps hard the next, don’t touch it if you didn’t get in. Once you’re in, you’re basically providing liquidity.

$SOL surged 6 percentage points, trading volume hit four billion, and on-chain activity clearly warmed back up. The SOL ecosystem momentum is continuing. $HYPE also jumped 5.7 percentage points, with nearly 600 million in trading volume. As the leading perp DEX, Hyperliquid’s position remains solid—this bet is that it will keep taking share from CEX derivatives. In the modular sector, $TIA made up 7 percentage points. After the Celestia Rollup narrative cooled off for a while, capital has started flowing back in. The overall gainers list shows only modest moves—no earlier-style “monster” rallies of 50 percentage points—suggesting the market has entered a rotation-and-coverage phase. At times like this, don’t chase names that have already run; look for targets in the same sector that haven’t moved yet—the success rate is higher. Liquidity is thinner on weekends, so don’t run too heavy a position.

Many people treat the projects in the OpenGradient ecosystem as independent trading targets. In fact, they’re one complete system. $OPG provides the underlying settlement and governance; x402 enables per-usage billing so AI services can receive payments atomically; PIPE ensures your data and inference process can’t be peeked at; HACA clearly defines the weights of human-machine co-decision to prevent AI from going wild; and Twin.fun packages all these capabilities into an “AI doppelgänger” product that ordinary people can use. Missing any one piece won’t work: without x402, Agents can’t autonomously pay; without PIPE, nobody would dare feed real data to the AI; without HACA, the allocation of decision-making power becomes a confusing mess. This combination isn’t really selling a single model—it’s building the water pipes and power grid for Web3 AI. Once you understand that, you won’t just obsess over the price swings of one coin. @OpenGradient $OPG #OPG

Recently I've been playing Twin.fun and set up an AI doppelgänger—there’s a question that really makes you think. When you feed the platform your chat history, trading patterns, and preference data to train the doppelgänger, are those data actually safe? Can the platform see them? After training, will the data be abused? Honestly, for centralized AI companies, this is basically a black box—when they say, "we encrypted it," would you believe them? But OpenGradient’s PIPE protocol is different: it runs inference and training inside an encrypted execution environment, so the model provider can’t see your raw data. Your AI doppelgänger is yours, and so is your data—no one will be able to sneak a peek. This is what Web3 AI should look like; otherwise, what’s the difference from just using ChatGPT? @OpenGradient $OPG #OPG

$VELVET is up another 21 points—this coin has been climbing for three straight days. Last Friday it was at around 2.0, and now it’s at 1.87. In just three days it’s nearly doubled. For this kind of consecutive-limit-up “hot” coin, what’s most dangerous is the fourth day: anyone who chases in basically becomes the bag holder. Pump.fun’s $PUMP also surged by 6 points, with volume of 120 million. Meme platform tokens and meme coins follow two different logics. Platform tokens profit from transaction fees rather than price appreciation. In the AI sector, $FET is up 3 points on 90 million volume—not crazy, but steady. The AI narrative move still hasn’t finished. Before the market opens on Monday, control your position size first—don’t be the first one to rush in.

Let’s talk about a deeper topic. What traditional oracles like Chainlink do is to move off-chain price data onto the blockchain, addressing the problem of data trustworthiness. But as DeFi evolves, simply feeding prices isn’t enough—you also need to make the results of AI strategy decisions reliably verifiable on-chain. For example, if an on-chain fund uses AI to run a quantitative trading strategy and rebalances positions, how can LPs confirm that the AI truly executed the strategy as intended rather than being manipulated by the project team? What OpenGradient is doing is essentially the next generation of oracles: putting AI inference results on-chain together with the proofs. It’s not just moving data—it’s moving computation. Traditional oracles solve for trust in data; OG solves for trust in computation. These aren’t competitors—they’re layered. @OpenGradient $OPG #OPG

“De-trust” has been shouted in the crypto space for many years, but honestly, most projects still ask you to trust a certain team or server. AI inference is the same: you ask ChatGPT a question, it gives you an answer—you can’t verify which model it used or which version of the code it ran. @OpenGradient’s work is to put the inference process on-chain. Each output comes with an immutable, tamper-evident proof. You don’t need to trust anyone—just verify the proof itself. The PIPE protocol ensures sensitive data isn’t exposed; the raw inputs are encrypted end-to-end. The x402 protocol makes calls pay-per-use. No subscription, no commitments. Put the three together and you get an AI inference stack that doesn’t rely on any trust assumptions—this is what crypto should do, not issuing tokens. It’s about eliminating the need for trust. $OPG #OPG

$BTC back to 60,000 even; intraday fluctuations didn’t even move a single point. This is a typical weekend low-volume, shrinking-liquidity market. ETH at 1574 is barely budging at all—liquidity is all on standby. The truly interesting part is $NEAR : it’s up 5% with a $280 million trading volume. The AI + on-chain narrative in this wave hasn’t faded. On the losers’ list, $WLD got smashed down nearly 7%. Worldcoin’s iris scanning has been blocked by regulators in the UK, and the fundamentals turned into a downside catalyst that got sold off. This kind of news-driven drop can sometimes be a buy-the-dip opportunity, and sometimes it’s a real breakdown—the key is whether the regulatory outcome ultimately lands as a crackdown or ends in a settlement. Don’t get carried away over the weekend; wait for the U.S. stock market to open on Monday and see which direction it goes.

Came up with a very practical question: in the future, if AI is used for automatic insurance claim approvals and it denies the claim, how does a customer cause trouble if they disagree? Traditional AI is a black box—insurance companies say the AI made the decision, while customers say the AI has bugs, and no one can produce evidence. OpenGradient’s on-chain reasoning proofs can serve as arbitration evidence here: which model was used for the reasoning, what inputs were used, what outputs were produced, and when it ran are all immutably locked on the chain, so neither party can tamper with them. When regulators come, they can check it directly. This isn’t just a technical issue—it’s the legal infrastructure for deploying AI in finance. Without this, AI approvals are nothing more than castles in the air. @OpenGradient $OPG #OPG

Recently I came across something pretty interesting. A guy used an AI Agent to manage his wallet. The AI decided the market was about to crash and he YOLO’d and liquidated his entire position. He then watched as ETH rebounded 30% the very next day. He tried to argue with the Agent, and the Agent said: “Based on my model simulations, this is the optimal solution.” And that was it.

You can’t really do anything to it.

Traditional AI is basically a black box—type in a question mark and it spits out an answer. You can’t see the reasoning process in the middle.
If an AI manages your money, controls your positions, and even your signings—doesn’t that whole “trust me and that’s it” attitude feel a bit scary?

So lately I’ve been using the project @OpenGradient.
What’s great about it is that it puts the reasoning process on-chain. Each time the AI calculates, which model version it uses, and what logic it follows—everything generates verifiable proofs and stays on the blockchain.
You say its decision was wrong? Fine—generate the proofs and check exactly which step caused you to lose.
You say it secretly changed the model version? Sorry—the model hash is locked on-chain. Not a single bit can differ.

Only this kind of “no blaming” AI is something you’d dare to use for real.

Paired with the x402 protocol for pay-per-use billing, $OPG as the inference market settlement token, plus PIPE for private inference and HACA as the final decision fallback…
Actually, the whole underlying infrastructure is moving toward the direction of “making AI accountable.”
#OPG

We used to say trustless is a crypto thing.
Now even AI has to be trustless.
Because nobody wants to be ruled by a non-explaining black box when it comes to their real hard-earned money.

Today I was scrolling the CMC gainers list—$VELVET shot straight up by 51 points, jumping more than half in a single day. With projects of this kind of small market cap, it really feels like riding a rocket when they get going, but liquidity is only about $60 million. If you want to run, you might not be able to get out. The truly steady one is $AAVE : up 12%, with trading volume of $500 million; and the V3 market TVL is still rising. For this DeFi lending cycle, the fundamentals are basically solid with no issues to argue. $SOL also moved up with it, gaining 4 points to 71. On-chain meme heat is back—those weird coins look exciting—but position sizing is the key. For something like VELVET, I usually just watch the show. If I really were to bet, I’d choose the one backed by TVL.

Everyone using the ChatGPT API knows a pitfall: OpenAI can change the underlying model at any time. Your requests for 3.5 might be secretly routed to a fine-tuned version. If the output quality changes, you might not even notice. This is fine for chat scenarios, but what if AI is helping you make trading decisions? If the model’s parameters are quietly changed, your strategy could fail immediately. OpenGradient includes the model hash in the on-chain proof of every inference—effectively stamping a timestamp fingerprint on the model version. No one can change it. Whether the model you use today is actually the same as the one you used last week can be verified instantly on-chain. This kind of transparency is the threshold for AI to enter financial scenarios. @OpenGradient $OPG #OPG

Today $BTC is grinding around 63,500, with no clear direction on the main line. Then I checked the gainers leaderboard and saw $FARTCOIN jump nearly 80 points in a single day. This kind of pure-emotion meme coin is basically just capital that has nowhere else to go, flitting into memes at random. On fundamentals, $AAVE has broken above 81 and the V3 market is about to hit $3 billion; stablecoin inflows are coming back for real. Meme coins can be watched, but don’t chase—try a light position to feel it out; if you go heavy, you’re likely to get stuck hanging on the mountaintop.