Emerging Trends

Emerging Trends in Fraud & OSINT
Focus	2024–2026 developments
Detection side	Deep learning, Benford + ML, real-time monitoring
Attack side	AI-driven fraud, deepfakes, BEC, crypto
2025 scale	$4.4T illicit; $579.4B fraud & scam losses
Regulatory shift	2026 CTA/BOI reporting rollback
Related	Analytical Methods, Benford's Law, AML Law

This article surveys emerging trends in fraud, forensic accounting, and OSINT for 2024–2026. Two forces dominate: artificial intelligence, which is simultaneously the most powerful new detection tool and the engine of a new wave of attacks, and a shifting regulatory landscape, most notably the 2026 rollback of beneficial-ownership reporting under the Corporate Transparency Act. Throughout, the recurring professional caution is to keep a human in the loop: machine models surface leads, but a qualified investigator must interpret, verify, and testify to them.

1 AI and machine learning in fraud detection

Machine learning has moved from research to routine practice in fraud detection. A growing body of systematic literature reviews (SLRs) surveys deep-learning approaches — neural networks that learn fraud patterns from large transaction datasets — with recent surveys catalouging architectures, datasets, and open problems.^[1] These methods complement, rather than replace, classical analytical procedures: they excel at flagging anomalies at scale but require careful handling of class imbalance (fraud is rare) and explainability (a flag a court cannot understand is of limited value).

A notable hybrid trend combines the century-old Benford's Law digit test with machine learning. Benford's Law screens for unnatural digit distributions in reported figures; feeding Benford-derived features into ML classifiers improves detection of fabricated numbers, as recent work (Gepp et al.'s "Fraud Detection Triangle" and Benford + ML studies) demonstrates.^[2] The pairing exemplifies the wider pattern of augmenting well-understood, defensible forensic tests with statistical learning rather than discarding them.

2 The AI-driven attack surge

The same technology arms offenders. The 2026 Nasdaq Verafin Global Financial Crime Report (2025 data) found that 90% of surveyed institutions report increased AI-driven attacks, against a backdrop of escalating scale.^[3]

Metric (Verafin 2026, 2025 data)	Figure
Illicit funds flowing globally	$4.4 trillion (up $1.3T since 2023; ~19.2% CAGR)
Combined bank-fraud + scam losses, 2025	$579.4 billion
Institutions reporting more AI-driven attacks	90%

These figures mark a sharp rise from the 2024 edition (2023 data), which reported over $3.1 trillion in illicit funds and roughly $485–500 billion in fraud and bank-fraud losses.^[3] AI lowers the cost and raises the quality of attacks: generative models draft convincing phishing and pretext messages at scale, synthesise voices and faces, and help offenders defeat identity checks. The AFP Payments Fraud surveys corroborate the threat environment, with a large majority of organisations hit each year and business email compromise consistently ranked the leading vector.^[4]

3 Blockchain and cryptocurrency fraud

Cryptocurrency remains a major and evolving fraud surface — as both a target (exchange collapses and thefts) and a laundering rail. The recent case record (see Case Studies) — FTX, Celsius, and Binance's AML failures — illustrates the range, from outright misappropriation of customer funds to systemic compliance breakdown. On the detection side, the public and immutable nature of most blockchains is a forensic advantage: cryptocurrency tracing clusters addresses, follows funds to exchange off-ramps where identity can be compelled, and flags mixers and privacy tools. The trend is toward integrating on-chain analytics with traditional asset tracing and money-laundering analysis, so that fiat and crypto flows are examined as one continuous trail.

4 Deepfakes and business email compromise

The convergence of generative AI with business email compromise (BEC) is among the most consequential emerging threats. Traditional BEC — impersonating an executive or vendor to redirect a payment — is now augmented by deepfake audio and video, in which a synthesised voice or face of a trusted person authorises a fraudulent transfer, defeating the "call to confirm" control that once caught such fraud.^[3] Because BEC is repeatedly identified as the top payments-fraud vector, defences increasingly emphasise out-of-band verification through independently held contact details, callback procedures resistant to spoofing, and staff training that assumes voice and video can be faked. Deepfakes also threaten OSINT and anti-corruption work, where fabricated media can be planted to mislead investigators — reinforcing the discipline of verification.

5 Real-time monitoring and human-in-the-loop

As attacks accelerate, detection is moving from periodic review toward real-time monitoring: models score transactions as they occur and hold or flag suspicious activity before funds leave. This is a direct response to speed — instant payments and crypto settle in seconds, compressing the window for intervention. Yet the profession's consistent caution is to keep a human in the loop. Automated models generate false positives, can embed bias, and cannot exercise judgment about intent or context; a qualified investigator must review alerts, decide on escalation, and ultimately stand behind conclusions in a report or as an expert witness. Under the 2023 amendment to Federal Rule of Evidence 702, courts scrutinise whether an expert's opinion reflects a reliable application of a method to the facts — a standard that an unexplained black-box output cannot meet on its own.^[5] The trend is therefore not full automation but augmentation: machines for scale and speed, humans for judgment and accountability.

6 The 2026 CTA / BOI rollback

A major 2026 regulatory shift concerns beneficial-ownership information (BOI) reporting under the Corporate Transparency Act (CTA). The CTA had required US companies to report their beneficial owners to FinCEN, a measure long sought by anti-money-laundering advocates to pierce shell-company anonymity. After litigation and an interim rule (26 March 2025), FinCEN issued a final rule on 11 August 2026 (effective 14 August 2026) that permanently removes BOI reporting for all US companies and US persons; only foreign entities registered to do business in the US remain reporting companies (and only as to their non-US-person owners), and FinCEN is deleting previously collected US-person BOI.^[6]

The practical consequence for investigators is significant: a domestic beneficial-ownership registry that might have accelerated asset tracing and anti-corruption due diligence will not, for US entities, exist. The rollback shifts weight back onto traditional investigative methods — OSINT, public records, the net-worth method, and Bank Secrecy Act reporting (CTRs and SARs) — to identify hidden ownership. It is a reminder that the tools available to fraud investigators depend heavily on a policy environment that can move in either direction.

7 See also

See also: Analytical Methods, Benford's Law, Cryptocurrency Tracing, AML Law, Case Studies, Evidence and Testimony.

References

1. Deep-learning fraud-detection systematic literature reviews. 2025. arXiv:2502.00201; arXiv:2506.10842.
2. Gepp, A. et al. "Fraud Detection Triangle" (Benford + ML). Accounting & Finance, 2024. DOI 10.1111/acfi.13192; Benford + ML (ODU, 2025).
3. Nasdaq Verafin. "Global Financial Crime Report" 2026 ed. (2025 data): $4.4T illicit; $579.4B losses; 90% report more AI attacks. verafin.com. 2024 ed.: $3.1T epidemic.
4. AFP. "Payments Fraud and Control Survey" 2025/2026 (BEC #1 vector). financialprofessionals.org.
5. Federal Rule of Evidence 702, 2023 amendment (eff. 1 Dec 2023) — reliable application standard. law.cornell.edu.
6. FinCEN. "Beneficial Ownership Information Reporting Requirement Revision," final rule, 11 Aug 2026 (eff. 14 Aug 2026). federalregister.gov.