Introduction

In cross-border asset tracing, open-source intelligence (OSINT) and human intelligence (HUMINT) play complementary roles: one unlocking publicly available information at scale, the other providing discreetly acquired insight unavailable from public data alone.

Artificial Intelligence (AI) increasingly supports this workflow in both arbitration and court-judgment enforcement, accelerating discovery and analysis while human judgment remains essential. 

This article examines how OSINT and AI can improve speed and scale, why HUMINT remains important, and how the three approaches can support a coordinated investigative strategy.

How Do Investigators Navigate Complex Jurisdictions?

Identifying target assets is central to any recovery strategy, but the process is rarely straightforward. Asset tracing requires investigators to navigate legal (including legal recognition of foreign judgments or arbitral awards), procedural, informational, and cultural differences that vary sharply by jurisdiction. These differences can affect both the viability of a strategy and the outcome of an investigation. 

In some regions, records are outdated, inaccessible, or are still paper based. In others, access is technically available but structurally restricted. The following six jurisdictions illustrate common challenges, from missing records to restricted searches:

• France has no central land registry. Property records are held across more than 140 local registry offices, each with its own geographic jurisdiction and no interconnectivity between them. Although name-based searches are possible, there is no single nationwide search. Requests must be submitted to the relevant local offices, so investigators generally need a reasonable idea of where a target may own property to keep the search practical and cost-effective.¹
• Portugal generally prohibits searching the land registry by owner name alone; a search requires the property's registration number, parish, and municipality. This restriction is rooted in the right to privacy granted by the Portuguese Constitution. Courts and judicial authorities are the exception.² Separately, Portugal's beneficial-ownership register moved from unrestricted public access to a "legitimate interest" standard under a 2025 decree, following the CJEU's 2022 ruling – adding a further layer of process for investigators.³
• UAE (Dubai) makes it straightforward to verify who owns a specific property using title deed details, with in-person verification requiring an Emirates ID or passport. What is not readily available is the reverse: a public third-party search that returns everything a named individual owns. For investigators building an asset profile from a name rather than a known address, this is the real constraint.⁴
• Cyprus’s company registry can indicate that filings have been submitted but are still pending processing, meaning the public record may not yet include a company’s most recent information. Investigators may therefore need to supplement the online search by requesting certified records or seeking assistance from a local representative.⁵
• Nigeria’s land market remains overwhelmingly informal. Over 90% of land in Nigeria is unregistered. Nigeria’s Minister of Housing and Urban Development said in 2025 that less than 10% of land has been formally registered in the roughly 140 years since registration began, reflecting the cost, complexity, and delays involved in the formal process and helping to explain the continued reliance on informal, unrecorded property transactions.⁶
• Hong Kong SAR, China, began tightening access to land and company registry information in 2021. Third-party name-based property searches now require applicants to provide identification and, when searching another person’s holding, written authorization from that person, effectively closing off the reverse, name-first search investigators typically rely on unless the subject cooperates.⁷ A parallel phased regime progressively withheld directors' usual residential addresses and full identification numbers from public inspection. Separately, beneficial ownership information held in Significant Controllers Registers is not open to general public inspection.⁸

These six jurisdictions illustrate that the barrier is rarely a single, uniform problem. It can be a matter of limited formal registration (Nigeria), fragmented infrastructure (France), privacy restrictions (Portugal and Hong Kong SAR, China), a one-way system (Dubai, UAE), or a processing lag (Cyprus). 

Offshore jurisdictions add a further layer of complexity. Shares in offshore entities can be held in the names of nominees, so identifying the ultimate beneficial owner (UBO) behind a structure requires detailed tracing to establish the ownership and control relationships among individuals and entities, rather than relying on a single registry search.⁹

In this context, detailed tracing refers to the multi-step investigative process of following the ownership chain through each layer of a corporate structure until the investigation identifies the individual or individuals who ultimately own or control it, or establishes that the ultimate owner is a state or public entity. 

Why Is a Single Search Often Not Enough?

A registry search (say, a Companies House or offshore corporate registry lookup) may show only the immediate legal shareholder of record. In offshore structures, the registry result is very often:

• A nominee shareholder, someone (or a corporate service provider) holding shares on paper for someone else under a nominee agreement, while the real owner's interest is documented separately (or not documented at all in public records).
• A holding company in another jurisdiction, which itself may be owned by another holding company, and so on.
• A trust, where the registered "owner" is a trustee, but the real economic beneficiaries are the trust's beneficiaries.
• A nominee director who signs documents but may have little or no decision-making power.¹⁰

If the investigation stops at the first name, an investigator may end up identifying a professional trustee in the British Virgin Islands or a shell company in Cyprus, not the actual person controlling the assets.

How Do Investigators Conduct Asset Tracing?

Asset tracing is often an iterative process of building a picture across multiple layers, jurisdictions, and sources, especially when the underlying structure has been deliberately designed to obscure who really controls or benefits from an asset. Investigators typically draw on a combination of public records, leaked or litigation databases, and control analysis to work back from a target asset or company to the individuals ultimately behind it. The process generally involves the following steps:

• Mapping each layer: identifying every entity in the chain from the target asset/company up to the top, jurisdiction by jurisdiction.
• Distinguishing legal owners from beneficial owners at each layer: checking nominee agreements, declarations of trust, powers of attorney, or side letters that reveal who really controls or benefits from an entity, even though someone else's name is on the register.
• Cross-referencing multiple sources: company registries, publicly accessible leaked datasets (e.g. Panama Papers-type leaks), litigation databases, court filings, filings with the U.S. Securities and Exchange Commission and other regulators, UBO registers where they exist, and sometimes property or banking records where lawfully available.
• Following control indicators, not just ownership percentages: someone can be a UBO through voting rights, the right to appoint directors, or contractual control, even without owning shares directly.¹¹
• Corroborating identity: confirming that a name found in one jurisdiction's filing is the same real person found elsewhere (same date of birth, address, and passport details), since aliases, similar names, or slightly different spellings are common obstacles.
• Following the money, not just the paperwork: sometimes ownership documents are incomplete or deliberately opaque, so tracing also involves looking at who actually receives dividends, loan repayments, or distributions.

The result of this tracing is an ownership and control diagram that connects the offshore entity all the way back to one or more identifiable natural persons (the UBOs) or, where applicable, a state or public entity, supported by evidence at each link, rather than an assumption based on a single, possibly incomplete, registry entry.

Methodology: A Complementary OSINT and HUMINT Model

The jurisdictional patterns above point to a broader principle: OSINT is typically the precursor to HUMINT, not a substitute for it. Reviewing asset registries, litigation and regulatory records, and deep-web sources, and conducting financial and compliance checks, lay the groundwork and identify leads. But OSINT has real limits. Determining true ownership from recorded data alone can be inconclusive, particularly where records are unavailable, restricted, or unreliable.

This is where investigators turn to HUMINT: skilled, discreet, legitimate engagement with well-placed individuals to close the gaps data-driven methods cannot. Human-source inquiries are often most valuable early in a case, providing background and context on what happened and why. They can also help pinpoint the physical location of assets, support manual document retrieval from registries, and inform source-of-funds inquiries. Well-executed HUMINT can sharpen the overall picture of ownership.

How Can AI Enhance the Investigative Process?

Alongside HUMINT, artificial intelligence is now another force reshaping how investigators close these information gaps, not by replacing human judgment but by extending the reach and speed of the OSINT phase that precedes it.

The power of generative AI lies in its ability to process natural language (that is, everyday human language) at speed, identify patterns according to the instructions it is given, and produce output in natural language that people can readily understand. Underneath, these models work by predicting one token at a time based on patterns learned from vast amounts of training data, rather than by independently retrieving or verifying facts from a trusted source. Because fluent, confident-sounding output is not the same as accurate output, any AI-generated findings, particularly any "links" established between individuals or entities, must be carefully reviewed to confirm that those links are genuine and that all supporting information is properly sourced and independently verified.¹²

An AI-driven investigative framework can draw on OSINT, beneficial ownership registries, data from offshore leaks, and property databases to help identify hidden assets. Previously established techniques such as network analysis, anomaly detection, and natural language processing allow investigators to surface relevant documents, suspicious asset flows, and ownership structures that would be difficult and time-consuming to detect manually.

From Generative AI to Agentic AI

AI has the potential to increase speed and efficiency in asset tracing investigations. A meaningful shift is now underway from generative AI tools that answer a prompt to agentic AI systems that can autonomously execute multi-step investigative workflows, adapting as they go rather than waiting for the next instruction. 

This distinction matters for asset tracing specifically, because much of the work, including UBO tracing, sanctions and adverse media screening, and cross-referencing entity relationships across registries, is itself a multi-step process that agentic systems are increasingly built to run end-to-end. 

Examples of how agentic AI can be applied: 

• Multi-agent orchestration across the investigative pipeline: rather than a single tool performing a single task, specialized agents can each handle a stage of the investigation and hand off to one another. For example, one agent flags an anomalous pattern, a second autonomously researches entity relationships and beneficial ownership structure, a third scans open-source and regulatory publications for adverse media, and an orchestration layer assembles the outputs into a single risk picture.¹³
• Autonomous UBO and entity-structure tracing: where access is permitted, and not behind a paywall, agentic platforms can connect to a large number of business registries, sanctions lists, court dockets, and adverse media sources simultaneously, tracing beneficial ownership chains and cross-referencing sanctions exposure in a fraction of the time a manual, analyst-led review would take.¹⁴
• Purpose-built crypto-tracing agents: for on-chain assets, dedicated agents can follow the movement of funds across multiple blockchains and cross-chain bridges, tracking mixing activity and token swaps as an investigation unfolds, rather than running a single fixed query.¹⁵

In practice, agentic systems can handle routine review at scale, but they introduce reliability risks. Chief among these are hallucinations, which can cascade across multi-step or multi-agent workflows if not constrained by governance and verification. Human authorization at defined decision points is therefore a critical control, particularly in complex cases involving unusual jurisdictions, multi-layered structures, or ambiguous ownership. 

Courts and regulators increasingly expect AI outputs used in legal or regulatory contexts to be verifiable, with appropriate audit trails and explainability.¹⁶ Any asset-tracing strategy incorporating agentic AI should therefore build in controls that (i) minimize hallucinations through retrieval-augmented generation, source grounding, and confidence thresholds; (ii) require human review of material findings and authorization of consequential actions; and (iii) preserve full audit trails so outputs remain accurate, explainable, and  assessed for evidential reliability.

What Are the Practical Limitations of AI?

Despite AI’s ability to process large volumes of information, determining true ownership remains difficult where the underlying data is incomplete or inaccessible to the system. Non-public or decentralized registries, and jurisdictions with manual, paper-based records, limit AI’s ability to seek new information. AI is strongest once data has already been obtained through legal process or public sources. 

In summary, proper use of AI accelerates processing but does not replace strategic planning, multi-jurisdictional coordination, or experienced investigators and counsel. Where records are limited or unreliable, both AI and OSINT become constrained, and establishing true ownership becomes manual, jurisdiction-specific work. This is precisely where HUMINT steps in.

Conclusion

Effective global and offshore asset tracing requires rapid, jurisdiction-specific action to identify and secure assets before they can be dissipated, particularly where disclosure standards and evidentiary rules differ. Regulatory reforms are strengthening beneficial ownership transparency, but offshore structures can still obscure ownership, and no single tool is sufficient on its own.¹⁷

OSINT, AI tools, and HUMINT, used together and paired with targeted legal mechanisms such as freezing orders, disclosure orders, and cross-border discovery, provide a comprehensive strategy needed for identifying hidden assets and ultimate beneficial owners in complex investigations.

References

[1] Fabien Cordiez, “France Land Registry Searches: Get Information on French Real Estate,” France Land Registry, accessed September 17, 2026.

[2] European Land Registry Association, “How to Get the Information: Portugal,” accessed September 17, 2026.

[3] Sam Green, “How Portugal’s New Rules Reshape Customer Due Diligence,” Zango, October 30, 2025.

[4] Emirates Government Services Hub, “Title Deed Verification in Dubai,” accessed September 17, 2026.

[5] Republic of Cyprus, Department of Registrar of Companies and Intellectual Property, “Electronic Search in the Business Entities Register,” accessed September 17, 2026; Elena Moretti, “Cyprus Company Registry Extract: Guide,” VLO Law Firm, January 11, 2026.

[6] Busola Aro, “Over 90% of Lands in Nigeria Unregistered, Says Minister,” TheCable, March 19, 2025.

[7] “Registry Search Rights Get Tight,” The Standard, October 26, 2021; “Conducting Search,” The Land Registry, Government of Hong Kong SAR, China, accessed September 18, 2026.

[8] “New Inspection Regime: General FAQs,” Companies Registry, Government of Hong Kong SAR, China, accessed September 18, 2026; “Significant Controllers Register: Overview,” Companies Registry, Government of Hong Kong SAR, China, accessed September 18, 2026.

[9] Financial Action Task Force, Guidance on Beneficial Ownership of Legal Persons (FATF, 2023), 4–5, 16, and 48.

[10] Financial Action Task Force, Guidance on Beneficial Ownership of Legal Persons (FATF, 2023), 4–5, 16–17, and 48–49; Financial Action Task Force, Guidance on Beneficial Ownership and Transparency of Legal Arrangements (FATF, 2024), 4–5.

[11] Financial Action Task Force, Guidance on Beneficial Ownership of Legal Persons (FATF, 2023), 16–21.

[12] National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1, July 2024, section 2.2, “Confabulation.”

[13] Henry Fosdike, “Agentic AI in Financial Services: From Hype to Governance,” SymphonyAI, May 29, 2026. 

[14] Variance, “Variance Raises $21.5M Series A to Transform Risk Workflows with AI Agents,” Business Wire, March 31, 2026.

[15] TRM Labs, “TRM Labs Launches Co-Case Agent: An AI Assistant for Every Crypto Investigation,” March 25, 2026.

[16] Courts and Tribunals Judiciary, Artificial Intelligence (AI) Guidance for Judicial Office Holders, October 31, 2025; Bar Standards Board, Guidance on the Use of Artificial Intelligence and Other Technologies, valid from May 18, 2026.

[17] Financial Action Task Force, “Guidance on Beneficial Ownership of Legal Persons,” March 10, 2023.