How ChainAware Is Doing for Web3 What Google Did for Web2

X Space #28 — ChainAware AI Agents and Roadmap for Individual Users. Watch the full recording on YouTube ↗ · Listen on X ↗

X Space #28 focuses on a single, practical question: what tools does ChainAware offer individual Web3 users right now, and where is the product roadmap heading? However, to answer that question meaningfully, Martin and Tarmo spend the first half of the session establishing the strategic framework that explains why these tools exist at all. The answer reaches back 20 years to the early Web2 era — when two technologies transformed the internet from a 50-million-user enthusiast network into the global economy it became. Those same two technologies are now needed in Web3, and ChainAware is building them.

The Web2→Web3 Parallel: Two Technologies That Drive Exponential Growth

Martin opens X Space #28 with a historical observation that sets the entire session in context: Web3 today is in exactly the same position Web2 was approximately 20 years ago. At that point, the internet had around 50 million users — predominantly technology enthusiasts, early adopters, and people comfortable with command-line protocols. The transition from 50 million to billions of users did not happen through better technology alone. It happened because two specific problems were solved.

The first problem was fraud. In early Web2, credit card data transmitted over the internet was routinely intercepted. E-commerce fraud rates reached 8–9% of total transaction volume. Consumers were genuinely afraid to transact online — and that fear prevented the ecosystem from growing. Payment processors eventually solved this by introducing predictive transaction monitoring: AI systems that analyzed purchasing patterns in real time and flagged transactions that deviated from expected behavior before they completed. The fraud rate collapsed, trust recovered, and the ecosystem began its exponential growth phase.

The second problem was user acquisition. Before Google invented AdWords, online advertising consisted of banner ads, roadside signs with URLs printed on them (Martin recalls literally seeing newspaper advertisements with website URLs), and mass broadcast campaigns with conversion rates so low that Web2 businesses could not become cash-flow positive. Google solved this by developing micro-segmentation using search and browsing history — targeting users with advertising precisely matched to their demonstrated intentions. User acquisition costs collapsed, conversion rates rose, and Web2 businesses finally had sustainable economics.

The Invisible Hand Was Google

Tarmo draws a memorable parallel to economics: business schools teach the concept of the “invisible hand” — the market mechanism that matches buyers with sellers — without ever fully explaining what it is. In Web2, the invisible hand turned out to be Google’s ad technology. It created the matching layer between supply and demand that made the Web2 economy work at scale. Web3 currently lacks this matching layer entirely. Furthermore, Web3’s fraud problem is even more acute than Web2’s was, because blockchain transactions are irreversible — there is no credit card chargeback mechanism, no dispute resolution, no reversal. When fraud happens in Web3, the loss is permanent. For more on how ChainAware applies this framework to Web3 businesses, see our AI agents for Web3 businesses guide.

The Web3 Fraud Crisis: 7–8% of TVL Lost Annually

The scale of Web3’s fraud problem is not well understood even within the industry. Tarmo provides the calculation that puts it in perspective: the annual “hackers fee” — smart contract exploits, protocol hacks, and direct theft — amounts to approximately 2–3% of Total Value Locked (TVL) across all blockchains. Adding impersonation scams, direct fraud, social engineering attacks, and rug pulls brings the total to approximately 7–8% of TVL annually.

This figure is strikingly similar to Web2’s pre-fraud-detection era. In early e-commerce, approximately 8–9% of all online transactions involved fraud. The parallel is not a coincidence — it reflects a structural reality: when a financial ecosystem lacks effective fraud detection, bad actors expand to fill whatever space the absence creates. Consequently, the fraud problem in Web3 is not a temporary phase that the industry will naturally grow out of. It is a structural deficit that requires specific technological intervention to close.

Moreover, the human cost extends beyond the financial numbers. Tarmo emphasizes that when new Web3 users get defrauded or rug-pulled in their first interactions with the ecosystem, they leave and never return — and they tell others to stay away. Every fraud victim is a permanent loss to the ecosystem’s potential user base. Reducing fraud is therefore not just about protecting existing users — it is about creating the conditions under which new users can enter Web3 and stay. For more context on how this plays out in DeFi specifically, see our DeFAI guide and our complete KYT and AML guide for DeFi.

Why Conventional AML Systems Fail in Web3

Before explaining what ChainAware does, Martin and Tarmo explain why existing approaches to Web3 fraud prevention are insufficient — and why the word “prevention” rarely applies to them at all.

Most existing “fraud detection” systems in Web3 are, in Martin’s precise terminology, “documentation systems” or “accounting technologies.” They maintain databases of known bad addresses — wallets that have been associated with confirmed hacks, scams, or other fraud events. When a new address interacts with a platform, these systems check whether the address appears in the database. If it does, the address is flagged. If it doesn’t, the address passes.

The Documentation Problem

This approach has a fatal structural flaw in the Web3 context: blockchain transactions are irreversible. In Web2, documentation-based fraud detection was supplemented by reversal mechanisms — credit card chargebacks, payment holds, dispute resolution processes. If a fraudulent transaction slipped through the filter, the victim could often recover their funds. In Web3, there is no recovery mechanism. A transaction that completes cannot be undone. Therefore, the only fraud detection that provides real protection is forward-looking prediction — identifying fraud risk before a transaction occurs, not documenting that fraud occurred after the fact.

The Clean Wallet Problem

Additionally, documentation-based systems are trivially circumvented by sophisticated fraudsters. A bad actor simply funds a new wallet through a clean route — withdrawing from a centralized exchange like Binance — and starts with a completely clean on-chain history. The documentation database has no record of this new wallet. AML checks pass. The fraudster proceeds. As Martin describes: “You just go to the central exchange and route it out from the exchange to a new wallet. You start from there. You cannot lose AML there. It doesn’t work.”

Predictive AI solves both problems simultaneously. By analyzing behavioral patterns rather than address identity, it identifies fraud risk from the way a wallet behaves — not from whether its address appears on a list. A newly created wallet that begins behaving like a pre-fraud wallet (specific transaction patterns, interaction with certain contract types, timing signatures of known attack preparation) receives a high fraud probability score regardless of whether it has any prior history. For a detailed comparison of these approaches, see our article on forensic vs AI-based crypto analytics and our guide to crypto AML vs transaction monitoring.

ChainAware Predictive Fraud Detector: 98% Accuracy, Free to Use

ChainAware’s fraud detector launched publicly on February 4, 2023 — exactly two years before X Space #28 was recorded. At the time of the session, Martin notes that BNB Smart Chain had just been announced and retweeted by BNB Chain’s official account (3+ million followers), marking a significant expansion of the product’s chain coverage.

The tool is straightforward to use: enter any wallet address, and receive a real-time fraud probability score. Regular addresses process in 0.5–1 second. Larger addresses with extensive transaction histories (Martin uses Vitalik Buterin’s ETH address as a demo benchmark) take slightly longer due to the volume of transaction data being analyzed — but still return results in real time.

The 98% accuracy figure requires context to be meaningful. It is not a self-reported claim — it is backtested against CryptoScamDB, an independent public database of confirmed crypto scam and fraud addresses. The model was tested against labeled data it had never seen during training (to prevent overfitting, also known as “curve optimization” in quantitative finance terminology). Of wallets the model identified as high fraud probability, 98% matched confirmed fraudulent addresses in the ground-truth dataset.

Chain Coverage

At the time of X Space #28, fraud detection covers Ethereum (the first chain supported), BNB Smart Chain (just announced), Polygon, and TON. Base is the next chain in active development, driven by client demand. Martin explains the prioritisation logic: “When clients are screaming for Base, of course we format to the clients.” Additional chains are added continuously as client demand warrants. For the current full list of supported chains, see chainaware.ai/fraud-detector.

Dynamic Detection — Not a Static List

One of the most practically important properties of ChainAware’s fraud detector is that it produces changing scores over time. Martin describes it explicitly: “You check some address, he’s good. Then you check him again some transactions later, he’s bad.” This reflects the model detecting behavioral pattern changes — an address that was behaving normally begins exhibiting pre-fraud behavioral signatures. The score changes accordingly, providing early warning of addresses that are transitioning from clean to risky behavior. Static AML databases cannot do this — an address either is or isn’t on the list, and that status changes only when a fraud event is confirmed and documented after the fact. For the complete methodology guide, see our Fraud Detector guide.

How the Predictive AI Models Actually Work

Martin provides a clear explanation of the model training process — important context for understanding why ChainAware’s approach is defensible and why competitors cannot simply copy it overnight.

The process begins with labeled training data in two categories. The first category is “positive behavior” — wallet addresses with confirmed histories of legitimate, clean activity across various DeFi protocols, exchanges, and use cases. The second category is “negative behavior” — wallet addresses associated with confirmed fraud, hack preparation, scam activity, and other malicious patterns, including the behavioral history those wallets exhibited in the weeks and months before their fraudulent events.

Training Takes Time — By Design

The model training process is iterative and time-consuming — not just because of computational requirements, but because improving a predictive model resembles learning mathematics: it requires repeated cycles of hypothesis, testing, refinement, and validation. As Martin explains: “It’s like learning mathematics. It’s a very iterative process.” ChainAware has been refining these models since before the public launch in February 2023, meaning the current production models represent over four years of iterative development.

Crucially, ChainAware builds its own proprietary neural networks. The company does not use OpenAI, DeepSeek, or any other third-party LLM provider for its prediction models. As Martin states explicitly: “We are not using OpenAI, we are not using DeepSeek. We have our own AI models, our own predictive AI models.” This is what creates the defensible competitive moat — not just the model architecture, but the years of labeled training data specific to blockchain behavioral patterns across multiple chains and millions of addresses. For more on why proprietary models matter, see our guide on attention AI vs real utility AI and our predictive AI for Web3 guide.

Google Cloud Partnership: The Road to 99%+ Accuracy

One of the significant announcements in X Space #28 is ChainAware’s acceptance into the Google Cloud Web3 Startup Program — an elite program providing compute credits to selected Web3 and AI startups. ChainAware received $250,000–$350,000 in Google Cloud compute credits, enabling a substantial expansion of its calculation capacity.

The practical implications are significant and specific. First, pre-calculation: rather than calculating fraud probability only when a query is received, ChainAware can pre-calculate scores for known addresses and update them continuously. This is analogous to the difference between on-demand rendering and cached rendering in software — faster response times, higher accuracy, and the ability to run more computationally intensive models.

The 99% Accuracy Target

Martin explains a specific trade-off the team encountered: a more data-intensive version of the fraud model (incorporating approximately 57 times more data per address) achieves 99%+ accuracy but at the cost of real-time performance. Instead of 0.5–1 second response times, the higher-accuracy model requires longer computation. With the Google Cloud compute credits, ChainAware can run pre-calculations that bridge this gap — computing the higher-accuracy scores in advance and serving them in real time when queries arrive.

Furthermore, the additional compute capacity opens the door to a capability that Tarmo describes as particularly valuable: timing prediction. Currently, ChainAware predicts whether fraud will occur — but not when. With sufficient compute power, the model can potentially predict the approximate timeframe of a fraud event. As Tarmo puts it: “We can even go over and start predicting when fraud is going to happen or when rug pull is going to happen.” This would transform the tool from a binary risk indicator into a temporal early-warning system. For context on how this fits into ChainAware’s broader product vision, see our guide to 5 ways Prediction MCP turbocharges DeFi platforms.

The Rug Pull Industry: Organised Crime With Social Psychologists

Before introducing the rug pull detector, Tarmo provides context that reframes how most people think about rug pulls. The common mental model is a small-scale scam — one bad actor creates a token, pumps it with fake activity, and pulls the liquidity. The reality is considerably more sophisticated and alarming.

Rug pulls are run by a professional industry. This industry employs social psychologists who design the social campaigns — the Telegram group strategies, the influencer playbooks, the FOMO messaging that convinces new users to buy. It employs engineers who design the rug pull contracts themselves, building in hidden mechanisms that allow the developer to drain liquidity at any chosen moment while appearing legitimate to cursory inspection. It employs marketing teams who manage the communication infrastructure. Tarmo describes it plainly: “It is huge industry and very profitable industry.”

The 95% PancakeSwap Statistic

The scale is illustrated by data ChainAware collected during a monitoring exercise: between 1,400 and 1,800 new liquidity pools are created on PancakeSwap daily. Of these, approximately 95% end in a rug pull. This is not a fringe phenomenon — it is the dominant outcome for new token launches on one of the largest DeFi platforms in existence. The 5% of legitimate new launches are effectively camouflaged within a sea of professionally engineered fraud.

Tarmo makes the decisive technical point about why static approaches to rug pull detection are inadequate: “You have to understand that your adversary is very well organised. They have very highly paid social psychologists, engineers. It is an industry, run very very well. And you have to answer to this fraud industry with behavioural analytics algorithms.” A static algorithm — one that checks contracts against lists of known rug pull patterns — is fighting a dynamic adversary with a fixed weapon. The rug pull industry simply adapts, creating new contract structures that bypass the known patterns. As Tarmo frames it: “One guy goes to war with an automated weapon, the other goes with a sword. It doesn’t work this way. If you respond with a static algorithm, you just lose.” For more on this dynamic, see our Rug Pull Detector guide and our guide to identifying fake crypto tokens.

ChainAware Rug Pull Detector: Predicting Before You Lose Everything

The rug pull detector operates on the same predictive behavioral principle as the fraud detector, but applies it to smart contracts rather than wallet addresses. Where fraud detection asks “will this wallet commit fraud?”, rug pull detection asks “will this contract execute a rug pull?”

The key distinction — which Tarmo emphasizes repeatedly — is that the tool predicts rug pulls, it does not document them. Documenting a rug pull is useful for reporting purposes but irrelevant to the investor who has already lost their funds. Prediction before the event is the only form of protection that matters given the irreversibility of blockchain transactions.

The practical implication is direct: before investing in any new token or liquidity pool, paste the contract address into ChainAware’s rug pull detector. The model analyzes the contract structure, the developer wallet’s behavioral history, the liquidity dynamics, and the trading patterns to produce a risk assessment. If the result indicates high rug pull probability, the investment decision is clear regardless of how compelling the project’s marketing appears.

Chain Coverage for Rug Pull Detection

At the time of X Space #28, rug pull detection covers Ethereum and BNB Smart Chain. Rug pull analysis requires somewhat more computation than wallet-level fraud detection — the contract analysis is more extensive — making chain expansion a more incremental process. Additionally, pre-calculation capabilities from the Google Cloud partnership will be applied to rug pull detection as well, increasing both speed and accuracy. For a full walkthrough of what the rug pull risk indicators mean, see our complete Rug Pull Detector guide.

Wallet Auditor and Share My Wallet Audit

The wallet auditor extends the fraud detector from a binary risk score to a comprehensive behavioral profile. When a user runs a wallet audit on any address, they receive not just a fraud probability but a complete picture of the wallet’s on-chain identity: experience level, risk willingness, behavioral intentions, protocol categories used, forensic analysis, and hundreds of individual attributes and summary metrics.

Martin explains the distinction between risk “willingness” and risk “ability”: “It’s the willingness to take a risk — slightly different. It’s like how banks are working. They are looking on your willingness to take a risk.” This distinction matters for behavioral prediction. Two wallets might have similar portfolio sizes (ability to take risk) but very different behavioral patterns — one consistently takes maximum leverage, the other consistently de-risks. The behavioral willingness metric captures this difference, making it a useful signal both for fraud assessment and for marketing targeting.

Share My Wallet Audit: Trust Without KYC

The most innovative feature of the wallet auditor is the Share My Wallet Audit capability. Web3’s pseudonymous nature creates a genuine verification problem: anyone can claim to be an experienced DeFi participant, a trustworthy counterparty, or a legitimate service provider. Without KYC, there is no identity verification mechanism. Consequently, fraud through impersonation is endemic — people claim expertise or trustworthiness they don’t have, or impersonate known legitimate actors.

Share My Wallet Audit solves this without requiring KYC. The process is simple: the wallet owner connects their wallet to ChainAware and generates a unique shareable link. Generating this link requires signing a transaction, which cryptographically proves that the person generating the link controls the wallet. The link recipient can then view the complete wallet audit — fraud score, experience level, risk profile, behavioral history, intentions — and verify that the audit belongs to the wallet the other party claims to own.

Martin draws the parallel to Web2 social verification: “In Web2, we have a word — ‘I Googled you.’ When I heard it first time, I was like… you what? But it’s like — trust but verify. Check the address.” The equivalent in Web3 is asking a business partner or counterparty to share their wallet audit before proceeding with any transaction or agreement. Furthermore, the person sharing cannot manipulate the result — the audit is calculated by ChainAware’s models from public on-chain data. It reflects actual behavior, not self-reported information. For more on how this applies to peer-to-peer and B2B contexts, see our behavioral user analytics guide.

AI Credit Score: Web3’s Missing Financial Layer

ChainAware’s credit score is the oldest component of the product suite — it predates the public launch of the fraud detector, having been developed first for SmartCredit.io’s DeFi lending platform. The credit score model has been running in production for over four years at the time of X Space #28.

Martin provides the clearest explanation of what a credit score actually measures — important context because the term is often misused: “Credit score displays the person’s financial ability to conform to, to respect his financial obligations. It’s their financial ability.” In traditional finance, FICO and similar scores are calculated from cash flow patterns, repayment history, debt levels, account longevity, and credit utilization. ChainAware’s Web3 credit score calculates equivalent metrics from on-chain transaction data — without any KYC, without any personal data collection, entirely from public blockchain history.

The credit score is currently less widely used in Web3 than fraud detection — Martin acknowledges that Web3 is “less focused on financial ability at the moment.” However, this is a temporary state of the market rather than a permanent feature. As DeFi lending matures and moves toward undercollateralized products, credit scoring becomes essential infrastructure. The protocols that build credit assessment infrastructure early will have a significant advantage when market demand catches up to the capability. For the full credit scoring guide, see our complete Web3 credit scoring guide and our DeFi credit score platform comparison.

Telegram Mini App, Bots, and Discord Integration

A significant practical convenience announcement in X Space #28 is the Telegram Mini App — which Martin notes was launched but not yet formally announced. The motivation is direct: Web3 users spend enormous amounts of time in Telegram groups discussing projects, sharing contract addresses, and making investment decisions. Previously, checking an address required leaving Telegram, navigating to the ChainAware website, pasting the address, and returning to Telegram with the result. This friction reduces the likelihood that users will actually check addresses before acting.

No Context Switch — Stay in Telegram

The Telegram Mini App eliminates this friction entirely. Users can paste any wallet or contract address directly into the TMA’s interface without leaving the Telegram environment. The result appears immediately within Telegram, making fraud and rug pull checks effortless at the exact moment they are most relevant — when an address is being shared in a group conversation. As Martin describes: “No context switch. Super effective.”

Additionally, ChainAware supports TON chain fraud detection — directly relevant to Telegram users given Telegram’s deep integration with the TON ecosystem. The timing is significant: Telegram updated its terms and conditions in early 2025 to require that all Telegram Mini Apps support TON chain. Martin notes this with evident satisfaction: “For us it’s like — thank you guys. Now only the real TMAs will stay there.” This requirement filters out the many TMAs that are unrelated to Web3 and ensures that genuine blockchain tools like ChainAware’s TMA remain prominent and accessible. ChainAware also offers Telegram and Discord bots with command-line interfaces for users who prefer typed commands to graphical interfaces.

Decentralised Ecosystem Self-Cleaning: The Big Picture

One of the most compelling ideas in X Space #28 is the concept of decentralised ecosystem self-cleaning through widespread adoption of free fraud detection tools. The logic is elegant and worth laying out explicitly.

Currently, most Web3 users check addresses only occasionally — when they are already suspicious, or after they have had a bad experience. Bad actors thrive in this environment because the overwhelming majority of their potential victims do not check. Consequently, the expected cost of attempting fraud is low: the probability of being caught before the fraud occurs is minimal.

If, however, checking addresses becomes a standard behavior — as automatic as Googling someone before a meeting in Web2 — the dynamic reverses entirely. Every bad actor’s address is continuously scrutinized. High fraud probability scores become visible to potential counterparties before any interaction occurs. The bad actor gets excluded from interactions not by a central authority but by individual users making informed decisions.

Martin describes the mechanism: “If you allow any user to check anyone — is it bad or good — you are introducing a fully decentralised system to verify users. Users are verifying each other.” This is not blockchain surveillance or KYC compliance — it is a peer-verification system built on public data and open tools. Furthermore, the more people who use the free tools, the more effective the ecosystem cleaning becomes. Unlike most network effects where early adopters benefit most, this one benefits late adopters equally — every new user who starts checking addresses adds to the collective protection of the ecosystem.

Web3 Ad Tech: The Second Key to Exponential Growth

While X Space #28 focuses primarily on individual user tools, Martin and Tarmo repeatedly reference the second pillar of ChainAware’s vision: Web3 ad tech. This is covered in detail in X Space #29 and subsequent sessions, but its importance to the overall framework warrants explanation here.

The user acquisition problem in Web3 is severe. Martin describes a real client scenario: 3,000 monthly website visitors → 600 connected wallets → 6–8 transacting users. This 0.2% conversion rate makes Web3 user acquisition unit economics fundamentally unworkable. No business can become cash-flow positive acquiring transacting users at this conversion rate, regardless of how low the cost-per-click is.

The root cause is the same mass-broadcast approach that Web2 used before Google AdWords: every visitor sees the same message, regardless of who they are and what they want. Web3’s blockchain data makes this unnecessary — every connecting wallet brings a complete financial behavioral history that can be used to deliver precisely targeted messages. An experienced DeFi borrower and a first-time crypto user visiting the same lending protocol should see completely different messaging. Currently, they see the same thing.

Meme Coin or Cash-Flow Positive — The Binary Choice

Martin frames the strategic choice facing every Web3 project founder with characteristic directness: “Either you are a meme coin or you will become cash-flow positive. And to become cash-flow positive, first you have to eliminate fraud. Second, apply ad tech like Google for Web3.” There is no third option. Projects that continue using ineffective mass-broadcast marketing while failing to address fraud will exhaust their token treasury without achieving sustainable economics. For the full detail on how ChainAware’s marketing agents solve this, see our guides on the personalization opportunity in Web3, the DeFi onboarding problem, and our SmartCredit case study showing 8x engagement improvement.

Comparison Table: ChainAware Individual User Tools

Tool	What It Does	Chains Supported	Accuracy	Cost	Access
Fraud Detector	Predicts future fraudulent behavior from wallet behavioral history	ETH, BNB, MATIC, TON, BASE	98% (backtested CryptoScamDB)	Free	chainaware.ai/fraud-detector
Rug Pull Detector	Predicts whether a contract will execute a rug pull	ETH, BNB	High (ML-based)	Free	chainaware.ai/rug-pull-detector
Wallet Auditor	Full behavioral profile: experience, risk willingness, intentions, categories, forensics	ETH, BNB, MATIC, TON, BASE	Real-time	Free	chainaware.ai/audit
Share My Wallet Audit	Cryptographically signed shareable wallet audit link — proves wallet ownership + profile	ETH, BNB, MATIC, TON, BASE	Real-time	Free	Via Wallet Auditor
AI Credit Score	On-chain financial ability score — FICO equivalent for Web3	ETH	4+ years production	Free (individual)	chainaware.ai/credit-score
Telegram Mini App	Fraud + rug pull checks inside Telegram — no context switch	ETH, BNB, TON	Same as web tools	Free	Telegram search: ChainAware
Telegram Bot	Command-line wallet checks via Telegram message	ETH, BNB, TON	Same as web tools	Free	Telegram search: ChainAware
Discord Bot	Wallet and contract checks inside Discord	ETH, BNB	Same as web tools	Free	Discord integration

Frequently Asked Questions

Why is ChainAware’s fraud detection free for individual users?

ChainAware keeps individual user tools free because widespread adoption creates ecosystem-level value that benefits the company indirectly. When every Web3 user checks wallet addresses before transacting, bad actors get excluded organically — the ecosystem self-cleans. This increased trust and safety benefits all ChainAware clients, including enterprise customers who pay for the transaction monitoring agent and marketing agent. Additionally, free tools attract users who eventually become enterprise clients or refer enterprise clients. The infrastructure costs are real — ChainAware pays for the compute — but the Google Cloud partnership helps offset these costs substantially.

What is the difference between AML screening and ChainAware’s fraud detection?

AML (Anti-Money Laundering) screening checks whether a wallet address appears on lists of known bad actors — sanctioned entities, confirmed fraud addresses, mixer service users. It is backward-looking and documentation-based: it can only flag addresses that have already been identified as problematic. ChainAware’s fraud detection is forward-looking: it predicts whether an address will exhibit fraudulent behavior in the future based on its behavioral patterns, regardless of whether it has any prior documented history. In Web3 where transactions are irreversible, only forward-looking prediction provides meaningful protection. For the detailed comparison, see our AML vs transaction monitoring guide.

How does the Share My Wallet Audit feature prove wallet ownership?

When a user generates a Share My Wallet Audit link, they must sign a message with their private key — the same mechanism used to sign any blockchain transaction. This cryptographic signature proves that the person generating the link controls the private key associated with the wallet address. The link is then unique to that signing event and cannot be generated by anyone who doesn’t control the wallet. The recipient of the link can therefore be confident that the wallet audit they see corresponds to the wallet whose ownership the sender has proven.

What does ChainAware’s 98% accuracy actually mean?

The 98% accuracy figure is derived from backtesting against CryptoScamDB — an independent public database of confirmed crypto scam and fraud addresses. The model was tested on labeled data it had never seen during training (using held-out test sets, not training data, to prevent overfitting). Of all wallet addresses the model flagged as high fraud probability, 98% matched confirmed fraudulent addresses in the ground-truth dataset. Backtesting methodology is essential for any predictive AI claim — as Martin notes in the X Space, “an algorithm without backtesting is like a ship without a captain.” For more on ChainAware’s methodology, see our Fraud Detector guide.

Why does ChainAware use its own AI models rather than OpenAI or other LLMs?

LLMs (large language models) like ChatGPT are statistical autoregression engines designed to predict the next word in a text sequence. They are not designed for behavioral prediction from structured blockchain data — and cannot provide measurable accuracy for fraud detection tasks. Building proprietary neural networks trained specifically on blockchain behavioral data (labeled examples of pre-fraud and pre-rug-pull wallet behavior) produces models with verifiable, backtested accuracy that continuously improves as more labeled data accumulates. Using LLMs would provide no competitive advantage since any competitor could build the same wrapper. For the full explanation, see our predictive AI vs LLMs guide.

What is the Google Cloud Web3 Startup Program partnership?

ChainAware was accepted into Google’s Cloud Web3 Startup Program, receiving $250,000–$350,000 in compute credits. The program is selective — Google verified that ChainAware had real calculation needs rather than just wanting to make an announcement. With these credits, ChainAware can run pre-calculations that push fraud and rug pull prediction accuracy above 99%, reduce latency, and begin developing timing prediction capabilities (predicting when a fraud event will occur, not just whether it will occur). This compute partnership is a significant enabler of the roadmap milestones discussed in X Space #28.

How is this X Space series structured?

ChainAware co-founders Martin and Tarmo have been hosting weekly (previously biweekly) X Spaces since January 2024. X Space #28 focuses on individual user tools and roadmap. X Space #29, covered in our article on attention AI vs real utility AI, discusses the broader Web3 AI landscape. Subsequent sessions cover business-facing tools and DeFi AI applications — see our full series including the real AI use cases guide, the AI agents for Web3 businesses guide, and the DeFAI explained guide.

This article is based on X Space #28 hosted by ChainAware.ai co-founders Martin and Tarmo. Watch the full recording on YouTube ↗ · Listen on X ↗. For questions or integration support, visit chainaware.ai.