VisionX AI Launches AEGIS Large Graph Model for Financial Crime

VisionX AI launched AEGIS, a Large Graph Model for tracing criminal networks, and TraceX, a blockchain investigation product, but named no customers.

VisionX AI Launches AEGIS Large Graph Model for Financial Crime

VisionX AI, a Singapore RegTech company that operates as a standalone business within Alpha Ladder Group, launched on November 18 AEGIS, a Large Graph Model for financial crime, and TraceX, the first product built on it. AEGIS brings graph AI to financial crime detection, learning how accounts, wallets and entities connect and how money moves between them, while most monitoring systems still score payments one at a time.

For compliance teams at banks, the release puts a focus on how well their own financial crime detection, whether built on graph methods or AI, sees relationships rather than single transactions. VisionX describes TraceX as a product for investigators and law enforcement working on blockchain activity.

How AEGIS models transaction networks

AEGIS uses a Graph Transformer architecture and was pretrained on graph data drawn from financial transactions. A graph model treats wallets, accounts and entities as nodes and the transfers between them as links, so it can learn the structure of how funds travel instead of judging each payment on its own attributes.

VisionX pairs the model with large language models from established providers in a hybrid set-up. AEGIS supplies the network intelligence, while the language models handle conversation, reasoning and AI-supported investigative work. The company says the combination is meant to surface concealed relationships, unusual fund movements and the wider criminal groups operating behind single transactions.

The name stands for Attention-based Entity and Graph Intelligence System. VisionX took it from the shield of Zeus in Greek myth, which was lent to Athena.

Tin Pei Ling, executive chairwoman of VisionX AI and co-president of Alpha Ladder Group and MetaComp, framed the product around how crime is organized.

“Financial crime operates through networks of people, entities and financial activities. To fight it, we need AI that understands the relationships between transactions.”

Tin Pei Ling, Executive Chairwoman, VisionX AI

Graph analysis versus transaction-by-transaction monitoring

Most transaction monitoring in production today is rules-based. A scenario fires when a payment or a short run of payments crosses a threshold, such as an amount, a velocity or a geography, and an analyst then looks at the account behind the alert. The method is easy to explain to examiners, but it sees each alert largely in isolation.

That is a weakness against organized laundering, where a group splits funds across many accounts so that no single payment looks unusual. Each transfer can stay under a threshold while the pattern across the whole group does not. Network analytics looks at that pattern directly, by tracing which accounts share counterparties, devices or beneficiaries and how quickly funds move through them.

For investigators, the practical difference is the starting point. A transaction-level alert asks whether one payment is suspicious, whereas a graph asks which other accounts and cases connect to it. VisionX CEO Summer Yu described the problem in those terms.

“Individual cases often look unrelated, even when the same criminal network sits behind them.”

Summer Yu, CEO, VisionX AI

Graph methods are not new in financial crime compliance, and many banks already run link analysis on customer data. What VisionX is proposing is a pretrained AI model that learns those graph structures for financial crime detection instead of relying on hand-built queries and rules. Buyers would typically compare such a model with a tuned scenario library on how many networks it finds and how many false positives it produces. Regulators’ expectations for transaction monitoring in the US, UK and EU still center on whether a system is calibrated to a firm’s own risks, and a model has to show that it is.

What TraceX does for crime investigators

TraceX is an AI-driven platform for blockchain tracing and investigation that is being developed for financial crime investigators and law enforcement bodies. It is intended to follow funds across linked blockchain addresses, flag suspicious patterns that connect separate cases, and produce investigative leads backed by traceable evidence and human oversight.

Several of the features that would make the graph approach most visible are not yet released. Reverse tracing, MegaMap network analysis, pattern recognition and case association are on the roadmap. Investigators keep decision-making authority and use TraceX suggestions as guidance, and every decision is logged.

That logging matters for audit. A record of what the model suggested and what the analyst decided is the kind of evidence a supervisor or an internal audit team asks for when reviewing how an alert was closed.

VisionX’s target markets and data

VisionX is targeting financial institutions, digital asset firms and public-sector bodies, and says that spinning the company out as an independent business widens the pool of buyers it can sell to.

The wider group does give some context. VisionX began as an incubated project inside MetaComp, which is licensed by the Monetary Authority of Singapore (MAS) as a Major Payment Institution, and it now supplies compliance intelligence for MetaComp’s StableX Agentic Network, which debuted as the StableX Network in November 2025.

Its existing products show where the data comes from today. VisionX KYT pools blockchain risk data from several independent analytics providers covering 1 billion wallets across 25 chains, and Web2.5 TxM merges identity, behavioral and network signals to monitor transactions across traditional and digital asset markets. AEGIS sits on top of that data estate, which suggests its strongest use is in digital asset flows rather than in retail bank payments.

Where graph AI fits FATF and MAS expectations

The launch cites the Financial Action Task Force (FATF), which on July 16 published its seventh targeted update on the FATF Standards for virtual assets and virtual asset service providers. The update found that criminal groups use the cross-border reach of virtual assets to defraud victims, evade sanctions and launder money, and that funds move through stablecoins, cross-chain bridges, decentralized exchanges and mixers, where they blend with legitimate activity and become harder to trace and freeze. That is the pattern a link-based method is designed to follow, because the trail runs through many addresses rather than one. A firm applying a risk-based approach still has to show that its monitoring matches the risks it has identified, whichever technique it uses.

On the AI side, MAS published a consultation paper on guidelines on AI risk management on November 13, 2025. The proposed guidelines apply to all financial institutions in Singapore and cover oversight of AI risk, risk management systems, policies and procedures, controls across the AI life cycle, and the capabilities and capacity needed to use AI. The consultation closed on January 31, 2026 and proposed a 12-month transition period after issue.

MAS published a response to feedback on October 7, 2026, alongside its final guidelines. The response sharpened how the guidelines apply in proportion to the risk of each AI use. A bank that buys a graph model or a hybrid system like AEGIS would need to inventory it, validate it and be able to explain its outputs to the board and to supervisors. A bank using AI for graph-based financial crime detection would obtain that documentation from the vendor during due diligence.

What comes next for VisionX and TraceX

VisionX plans to apply the AI model’s graph intelligence to financial crime detection, fraud and scam prevention, compliance risk management and investigations. Others in the market are also moving toward machine learning and graph techniques, as in Hummingbird’s acquisition of Quantimon, and banks are testing AI agents for AML, fraud detection and risk monitoring. A hybrid design that keeps a specialized network model separate from general-purpose language models gives a bank two components to validate instead of one, which is a governance question as much as a technical one.

For teams already tuning scenarios, the nearer-term step is to compare what a network view would add to current alerts. Any graph model would face the same test as transaction monitoring tuning: thresholds are tested and alert quality is measured before and after a change.

The next TraceX releases on VisionX’s roadmap are reverse tracing, MegaMap network analysis, pattern recognition and case association, which would bring the graph AI behind AEGIS to case-level financial crime detection.