Private markets are built on trust, confidentiality and personal relationships. Those same qualities are becoming easier to imitate. As generative AI improves the credibility of synthetic identities, cloned voices and personalised impersonation, private-capital firms and wealthy investors face a risk that sits less in markets than in the infrastructure through which capital moves.
Executive Summary
- US consumers reported more than $12.5 billion in fraud losses in 2024, while UK identity fraud rose 9 per cent in the first half of 2026.
- The strongest evidence suggests AI is industrialising established fraud techniques rather than creating an entirely new fraud economy.
- Private equity, private credit and family offices are attractive targets because large transfers often pass through fragmented, trust-based operating structures.
- For UHNW families, identity integrity and transaction authentication are becoming part of generational wealth preservation.
- AI-specific institutional loss measurement remains limited, making disciplined distinction between attempted, suspected and confirmed AI-enabled fraud essential.
The New Weak Point in Private Capital Is Trust
Private capital has always depended on something public markets require less visibly: confidence in counterparties whose identities, intentions and instructions are often confirmed through relationships rather than continuous public scrutiny.
A limited partner receives a capital-call notice from a general partner it has known for years. A family-office principal speaks to a familiar adviser. A finance team receives account details from a fund administrator. A portfolio-company executive recognises the voice of a senior colleague. In conventional operating environments, familiarity helps establish authenticity.
Artificial intelligence is weakening that assumption.
Bloomberg’s August 2026 Going Private reporting highlighted growing concern among private-capital firms over impersonation, identity theft and spoofing directed at investors. The importance of the story is not that private markets have suddenly discovered fraud. Fraudulent payment instructions, phishing, fake capital calls and executive impersonation all pre-date the present AI cycle. The more consequential question is whether generative AI is making those established methods cheaper to create, easier to localise and substantially more credible.
That distinction matters because the available evidence does not support the proposition that every increase in financial fraud is attributable to artificial intelligence. It does support a narrower and potentially more durable thesis. AI is reducing the cost of producing persuasive social engineering.
For private equity investors, private credit managers, family offices and ultra-high-net-worth individuals, the resulting risk is unusually sensitive because the underlying transactions are often large, confidential and operationally complex. A fraudulent consumer payment may be damaging. A compromised private-market payment instruction can involve millions.
The deepest change, therefore, is not that artificial intelligence has invented financial fraud. It is that AI is weakening the traditional signals by which sophisticated investors decide whom, and what, to trust.
The Fraud Economy Was Already Enormous Before AI
Any serious analysis of AI fraud in private capital has to begin with an uncomfortable baseline. Financial fraud was already a major economic problem before voice cloning and deepfake video became widely accessible.
According to the US Federal Trade Commission, American consumers reported more than $12.5 billion in fraud losses during 2024, 25 per cent more than the previous year. Investment scams alone accounted for approximately $5.7 billion, up 24 per cent from 2023, while imposter scams generated roughly $2.95 billion in reported losses.
Those figures are significant, but they are consumer data. They should not be presented as evidence of private-equity or institutional losses.
The FBI’s Internet Crime Complaint Center provides another window into the broader fraud environment. IC3 reported approximately $16.6 billion in cyber-enabled losses in the United States during 2024. That figure uses a different scope and methodology from the FTC dataset, so the two numbers cannot simply be added together.
By 2025, IC3 data showed overall reported investment-scam losses exceeding $8 billion. Within investment complaints where an AI nexus was reported, losses exceeded $632 million. Businesses separately reported more than $30 million in losses from business email compromise cases involving AI.
The gap between more than $8 billion in overall investment-scam losses and the smaller explicitly AI-linked subset is analytically important. AI may be present in more incidents than victims recognise, but the evidence does not allow every investment scam to be relabelled as an AI scam.
The United Kingdom shows a similar tension between rapidly rising identity risk and limited AI-specific measurement.
Cifas recorded nearly 130,000 identity-fraud cases during the first half of 2026, an increase of 9 per cent. Identity fraud represented approximately 59 per cent of fraud-risk cases in its dataset, while bank accounts and plastic cards accounted for 68 per cent of the identity-fraud cases cited.
Investment brands are also being impersonated directly. The Investment Association recorded 478 reports involving investment-management firms being impersonated in the second half of 2024. Approximately 23 per cent of those attempts succeeded, generating around £2.7 million in consumer losses.
There is no basis for claiming that all 478 incidents used artificial intelligence. The data is valuable because it demonstrates that investment-firm impersonation is already a functioning fraud model. AI potentially makes that model more convincing.
The broader UK banking system recorded approximately £1.28 billion in fraud losses during 2025, according to UK Finance. The same report cited BioCatch data from nine UK financial institutions showing a 62 per cent increase in scam attempts across more than 100 million accounts. That dataset is vendor-derived and does not represent the entire UK financial system, but it illustrates the pressure financial institutions are experiencing.
AI is entering a fraud economy that was already immense. The institutional question is therefore not whether AI caused the underlying problem. It is whether AI changes the conversion rate between an attempted deception and an authorised transfer.
AI Changes the Economics of Social Engineering
Generative AI attacks one of the traditional inefficiencies of financial fraud: producing convincing deception at scale.
A conventional phishing campaign can be undermined by poor grammar, generic language, cultural errors or obvious inconsistencies. Large language models reduce those imperfections. They can produce polished emails, imitate professional tone, translate content and generate highly specific narratives at very low marginal cost.
The same economics extend beyond text.
Voice-cloning technology can reproduce enough of a person’s speech patterns to weaken confidence in telephone verification. Video deepfakes can imitate familiar faces. Generative systems can create synthetic photographs, documents and supporting identities. Automated reconnaissance can combine executive biographies, professional networks, social-media material, corporate announcements and other public information into a more personalised attack.
Interpol has warned that AI, large language models and related technologies allow more sophisticated fraud campaigns to be conducted without requiring advanced technical capability and at comparatively low cost.
That is the economic mechanism that matters.
The criminal does not necessarily need a new fraud strategy. The criminal needs a better version of an old one.
A business email compromise request can be written more convincingly. A fake investment representative can communicate fluently. A fraudulent capital-call notice can more closely resemble legitimate correspondence. An impersonator can potentially sound like the person whose authority is being borrowed.
AI can also reduce language barriers, allowing attackers to target victims outside their own linguistic or cultural environment with fewer obvious warning signs.
The result is a potentially higher-quality attack surface at much greater scale.
This is why the phrase AI impersonation financial fraud is more analytically useful than the broader idea of an entirely new AI crime economy. The technology’s strongest demonstrated advantage is efficiency. It improves production, personalisation and credibility.
For sophisticated investors, this creates a fundamental distinction between recognition and authentication. Recognising the language, voice or face of a trusted counterparty is no longer equivalent to proving identity.
Why Private Equity and Private Credit Are Attractive Targets
Private capital offers criminals an unusually compelling combination of transaction value, operational complexity and limited visibility.
The private-equity lifecycle alone creates repeated points at which valuable instructions move between counterparties. Fund marketing leads to investor onboarding. Investors complete KYC and AML checks, execute subscription documents and make commitments. Capital calls follow. Distributions eventually move in the opposite direction.
Between those points sit fund administrators, custodians, private banks, law firms, accountants, placement agents and other advisers.
Private credit adds another layer of transaction activity, including funding instructions, borrower payments and potentially complex financing structures. Co-investments, continuation vehicles, secondaries, NAV financing, portfolio-company acquisitions, escrow arrangements and M&A closing payments create further opportunities for high-value instructions to move across institutions.
Every handoff expands the authentication problem.
Private markets also function through confidentiality. Information is intentionally restricted. Ownership structures can be complicated. Transactions may be cross-border. Technology stacks may be fragmented. Manual processes remain relevant. Email still carries operational importance.
None of those characteristics implies weak governance. Together, however, they create a useful environment for impersonation.
GP impersonation is an obvious risk because general partners have legitimate reasons to request funds from limited partners. LP impersonation can matter where administrators or managers are instructed to change investor details or process transactions. Fund-administrator impersonation may exploit the administrator’s role as an operational intermediary. Fraudsters can also spoof placement agents, create fake fund websites or fabricate co-investment opportunities.
Private credit is exposed to similar weaknesses. Where counterparties expect bespoke payment instructions, time-sensitive funding or multiple intermediaries, the attacker’s objective is not necessarily to compromise the underlying asset. It is to compromise the process around the asset.
That difference is strategically important.
Traditional investment risk asks whether an asset has been correctly priced. Fraud risk asks whether money reached the intended destination in the first place.
For private-market participants, those two forms of risk now have to coexist within the same capital-preservation framework.
A Capital Call Is a Payment Instruction and That Makes It Valuable
Capital-call fraud illustrates the problem in its clearest form.
A legitimate private-equity fund periodically asks its investors to transfer committed capital. The investor expects the request. The fund may specify an amount, deadline and bank account. A fraudulent capital call exploits that expected workflow.
The technique existed before generative AI. ACA Global has previously highlighted phishing emails and scam capital calls targeting private-equity firms.
Artificial intelligence does not need to reinvent the structure. It can improve execution.
Language can become more polished. Formatting can become more credible. A message can be personalised to the recipient. Publicly available information may help an attacker time correspondence around genuine fund activity. Supporting communications can be generated quickly.
Bank-detail changes are particularly sensitive.
If a fraudulent notice merely asks an LP to transfer money to the bank account it already has on record, the attack achieves little. The economically valuable moment occurs when an attacker persuades the recipient that legitimate payment instructions have changed.
A spoofed administrator, GP finance executive or adviser may therefore attempt to introduce new banking details immediately before a high-value transfer.
The core weakness is not necessarily whether the document looks real.
It is whether the payment-verification process allows appearance to substitute for independent confirmation.
That is why known-number callbacks, out-of-band authentication, dual authorisation and controlled procedures for bank-detail changes become disproportionately important. They shift the verification question away from whether a communication appears authentic and towards whether the instruction has been independently validated.
In an environment of increasingly convincing synthetic content, that distinction may become one of the defining operational controls in private capital.
When Familiar Voices and Faces Stop Proving Identity
The most dramatic AI fraud cases matter because they demonstrate how easily established human cues can become unreliable.
In 2024, a finance employee at engineering group Arup’s Hong Kong operation joined a video conference that appeared to include senior colleagues. The participants were deepfake representations. Instructions delivered during the call resulted in approximately $25 million being transferred across 15 transactions.
Arup is not a private-capital firm. The case should not be presented as evidence of private-equity losses.
Its relevance lies elsewhere.
The employee was not merely deceived by an unusual email address or poorly constructed attachment. The attack reportedly created an environment in which the familiar appearance of senior colleagues reinforced the legitimacy of the payment instruction.
The implication for investment firms is direct. Visual recognition can no longer be treated as a sufficient authentication layer for sensitive transactions.
Voice carries the same vulnerability.
In 2019, years before today’s generative AI boom, an executive at a UK energy company transferred €220,000 after attackers reportedly mimicked the voice of the chief executive of the company’s German parent. The incident is useful because it shows that synthetic voice fraud is not a wholly new phenomenon. What has changed is accessibility and potential scalability.
By August 2026, major US money managers had reportedly faced voice-phishing attempts in which technology was used to imitate trusted voices and seek sensitive information or access. Two Sigma, Millennium Management and Citadel were among the firms identified in reporting.
The Two Sigma attempt did not produce a confirmed financial loss in the cited evidence. That distinction is critical. A sophisticated asset manager being targeted demonstrates attacker interest, but an attempted attack is not proof that institutional controls failed.
The Investment Association’s UK data require the same discipline. The 478 investment-firm impersonation incidents recorded in the second half of 2024 provide clear evidence that cloning legitimate investment brands is an established criminal technique. They do not establish AI involvement in every case.
These examples collectively narrow the credible conclusion.
A familiar face is useful information, but no longer sufficient evidence. A recognisable voice can support a relationship, but should not function as a cryptographic credential.
For high-value finance, identity increasingly has to be proved rather than perceived.
Family Offices Have a Different Kind of Attack Surface
Family offices face a related problem, but the operating environment can be more personal.
A large financial institution may employ separate treasury, legal, compliance, information-security and operations teams. A family office can manage significant wealth with a comparatively lean staff.
Decision-making may depend on relationships built over decades. A principal may communicate directly with a private banker. A chief investment officer may work closely with lawyers, accountants and trustees. Personal assistants, estate managers, household employees and family members may all sit somewhere within the broader information network.
The same structure that creates discretion and speed can make impersonation unusually sensitive.
High-value payment authority may be concentrated among relatively few people. Cross-border structures can involve trusts, foundations, holding companies and private investment vehicles. Transactions can extend beyond conventional securities to property, art, philanthropy, luxury assets and private aviation.
The criminal attack surface is therefore not limited to the investment portfolio.
Digital exposure also extends beyond office systems.
Public interviews, corporate announcements, professional profiles, executive videos, family photographs, travel information and social-media posts can provide material for targeted social engineering. Next-generation family members may contribute an additional public footprint without participating directly in investment decisions.
For a sophisticated attacker, this information can help construct context.
The objective might be to impersonate the principal, an adviser, a family member or a service provider. It might be to establish credibility before requesting a transaction. It might instead be to compromise an intermediary whose access to sensitive information makes a later attack more persuasive.
That changes how UHNW cybersecurity risk should be conceptualised.
Cybersecurity for ultra-high-net-worth families is not simply a question of protecting devices. It intersects with identity integrity, privacy, transaction governance, succession and counterparty management.
The relevant unit of protection becomes the network around the wealth, not merely the person who legally owns it.
This is particularly important in family office cyber risk management because the consequences of compromise can propagate across multiple structures. A compromised adviser may expose information about several entities. A successful impersonation may target liquidity rather than long-term holdings. Fraud recovery may be difficult even where the underlying portfolio remains intact.
For generational wealth, the operational perimeter can therefore be as consequential as the asset-allocation perimeter.
The Defence Is Moving From Recognition to Verification
Financial institutions are responding by building identity controls around signals that are harder to reproduce than a familiar voice or well-written email.
Behavioural biometrics are one example. Instead of asking only whether a user has entered valid credentials, systems can analyse behavioural patterns that may reveal unusual activity. Device intelligence examines the characteristics and history of the device being used. Transaction anomaly detection looks for behaviour inconsistent with previous activity.
Liveness detection and identity proofing attempt to strengthen onboarding and authentication where images or video are involved. Continuous and risk-based authentication can reassess risk after the initial login rather than treating authentication as a single event.
UK Finance has noted that AI and analytics can support dynamic, real-time risk assessments and that behavioural intelligence can identify subtle anomalies even where valid credentials are being used.
That matters because compromised identity is different from compromised credentials.
A criminal who has stolen a password may still behave differently from the legitimate account holder. A fraudulent payment instruction may be technically authorised through a compromised channel while remaining economically inconsistent with normal behaviour.
Technology, however, is only one layer.
Multi-factor authentication, passkeys, hardware security keys, payment whitelists, role-based access, transaction limits and secure portals can narrow the number of ways an attacker can convert deception into money.
Human governance remains equally important.
Independent callbacks to known numbers, dual authorisation for major transfers, cooling-off periods following bank-detail changes and out-of-band confirmation can interrupt attacks even where the original email, phone call or video appears completely convincing.
This is the central change in deepfake detection for financial services.
Institutions cannot assume that detection technology will identify every synthetic artefact. A stronger architecture is one in which successful impersonation does not automatically create transaction authority.
Platforms serving sophisticated private clients operate within the same trust framework. Bancara’s stated client-protection infrastructure, for example, includes multi-factor authentication, secure portals, audit trails, verified withdrawal channels and pre-approved protocols. These controls illustrate the broader direction of travel: transaction governance is becoming part of financial infrastructure rather than an ancillary technology function.
Fraud Risk Is Becoming a Private-Market Infrastructure Cost
The direct cost of financial fraud is straightforward. Money is stolen.
The more significant private-market consequence may be the infrastructure required to prevent that outcome.
Banks must invest in fraud analytics, behavioural intelligence, device monitoring and authentication. Administrators may need stronger verification around capital calls, distributions and changes to investor instructions. Managers may spend more on cybersecurity, compliance, training and incident response.
Cyber insurers face their own questions around social-engineering exposure, exclusions and pricing.
None of these measures are free.
Stronger verification can also slow transactions. A bank-detail change that once moved through email may require multiple confirmations. A large payment may require dual approval. Investor onboarding may demand additional identity checks.
The private-market industry therefore faces a trade-off between transaction efficiency and authentication certainty.
For the largest managers, higher security expenditure may be absorbable within an already substantial operational budget. Smaller firms may experience a proportionately larger burden if cybersecurity, digital identity and compliance become increasingly fixed costs of institutional credibility.
That raises the possibility of a security advantage accruing to scale.
It does not follow that smaller managers will inevitably consolidate. The evidence does not support that conclusion. But rising technology and compliance requirements could become another economic pressure favouring firms capable of spreading fixed operating costs across larger asset bases.
Trusted administrators and custodians may also gain strategic value.
Historically, operational due diligence could be treated as a supporting component of manager selection. In an environment of synthetic identity and payment fraud, investors may place greater weight on how capital instructions are authenticated, which counterparties hold transaction authority and how exceptions are governed.
Operational resilience can therefore become a competitive attribute.
There may also be structural demand implications for cybersecurity, identity-verification platforms, behavioural biometrics, fraud analytics, RegTech and secure-payment infrastructure. These are category-level observations, not recommendations of individual securities.
Regulation adds another layer of cost and uncertainty. Authorities across the United States, United Kingdom, European Union and Australia are examining overlapping issues around fraud enforcement, digital identity, AML, cybersecurity, AI governance and operational resilience. The UK’s Fraud Strategy 2026 to 2029 treats fraud as a national priority and emphasises collaboration and information sharing.
Questions of liability are less uniform.
Where fraudulent instructions appear authorised, disputes may arise over responsibilities among investors, banks, managers, administrators and other counterparties. Consumer reimbursement frameworks cannot simply be assumed to apply to institutional transactions. Contractual allocation of risk, cyber-insurance terms and the quality of internal controls may therefore become increasingly important.
The result is a new category of private-market operating expense: the cost of proving that an instruction is genuine.
For the Ultra-Wealthy, Cyber Resilience Becomes Capital Preservation
UHNW investors traditionally evaluate wealth architecture through familiar lenses: custody, manager diversification, banking relationships, liquidity, tax structures, jurisdictional exposure, succession and counterparty risk.
Identity integrity now belongs within that framework.
Consider counterparty concentration. A family may diversify investment managers while concentrating large payment flows through one bank, administrator or internal decision-maker. From a market-risk perspective, the portfolio may be diversified. From an operational perspective, the structure may still contain a critical point of failure.
Custody architecture raises a similar question. The concern is not merely which institution holds an asset, but how instructions affecting that asset are authorised and authenticated.
Private-fund exposure introduces another layer. Manager selection may increasingly encompass the quality of operational controls surrounding capital calls, distributions and bank-account changes. Due diligence on investment strategy can remain excellent while transaction governance is weak.
Cross-border complexity matters because each additional bank, adviser, trustee, administrator or entity creates another relationship through which sensitive information or payment instructions may move.
Succession adds further complexity.
Next-generation family members may inherit economic interests before they inherit institutional habits. Digital identities, social-media footprints and communications practices can become part of succession governance in the same way that investment education and legal structures already are.
Liquidity management is also relevant. Fraud tends to target assets that can move. An illiquid private-equity holding may be difficult for a criminal to appropriate directly. The cash waiting for a capital call is more accessible. The distribution being redirected to a fraudulent account is more accessible still.
For wealthy investors, this creates an important distinction between portfolio risk and transfer risk.
Portfolio analysis asks what can impair the economic value of an asset.
Transfer-risk analysis asks whether legitimate wealth can be diverted during movement between trusted parties.
In that environment, platforms and institutions serving sophisticated clients operate not merely as execution venues but as nodes within a broader trust architecture. Bancara’s positioning around institutional infrastructure, transparency and governed transaction channels fits this wider shift, but no financial platform can reasonably be treated as immune to cybercrime. Security remains a system of controls, verification and disciplined operating procedures rather than a guarantee.
For family offices, private banks and UHNW investors, the strategic conclusion is therefore not to treat fraud prevention as separate from wealth management.
Identity, custody, authentication and operational resilience are becoming components of capital preservation itself.
AI May Be Scaling Old Fraud, Not Inventing New Fraud
The easiest AI-fraud narrative is also the least precise.
Phishing is not new. Spoofing is not new. Executive impersonation is not new. Fraudulent wire instructions and business email compromise existed long before generative AI reached mass adoption.
Even synthetic voice fraud has documented precedent going back years.
The empirical limitation matters. Explicitly AI-labelled loss datasets remain small relative to total investment and fraud losses, and victims do not always know whether AI was used against them.
Media attention can therefore run ahead of measurement.
It is possible that AI is being used in a much larger percentage of attacks than reported data captures. It is equally possible that some incidents casually described as AI-enabled are fundamentally conventional social-engineering attacks with only modest technological enhancement.
Institutional investors also possess advantages that retail victims may not. Approval matrices, professional treasury teams, compliance departments, administrators and banking controls can make high-value organisations harder to defraud even where the economic reward for criminals is greater.
Standard controls remain effective against many AI-enhanced attacks.
A cloned voice cannot defeat a policy requiring an independent callback to a verified number. A deepfake executive does not automatically defeat multi-person payment authorisation. A perfectly written fake capital call cannot redirect funds where bank-detail changes require controlled verification.
Digitisation could therefore strengthen private-market security over time rather than merely weaken it.
The same AI that improves social engineering can support anomaly detection and fraud monitoring. Better digital identity infrastructure can reduce dependence on subjective recognition. Behavioural analytics can identify deviations invisible to human reviewers.
The most defensible thesis is consequently narrower than the most dramatic one.
AI has not clearly created an entirely new category of financial crime. It changes the marginal cost, scale, localisation and credibility of established attacks.
That alone is consequential enough.
The Next Private-Market Premium May Be Trust
Between 2026 and 2030, several outcomes remain plausible.
In a base case, AI-assisted social engineering continues to expand while banks, administrators and investment managers gradually improve authentication. Fraud becomes another persistent operating cost of private markets rather than a systemic crisis.
In a more aggressive scenario, voice, video and identity synthesis improve faster than verification controls. Financial institutions respond by imposing more restrictive payment procedures, stronger authentication and higher levels of manual confirmation.
Another path is possible. Defence technology may catch up.
Behavioural analytics, device intelligence, digital identity and stronger authentication could reduce the conversion rate of fraudulent attempts even as attack volume rises. More scams would be attempted, but realised institutional losses could remain comparatively contained.
Regulation could accelerate the same process. Authorities and financial institutions may formalise stronger verification expectations around sensitive transactions. The trade-off would be familiar: better security accompanied by greater onboarding and transaction friction.
The most disruptive scenario would be a major, publicly visible private-capital fraud exposing weaknesses in capital-call or transaction-verification infrastructure. Such an event could rapidly tighten LP due diligence, administrator responsibilities, insurance requirements and payment controls.
No numerical probabilities can credibly be attached to these scenarios from the available evidence.
What can be said is that trust is becoming more expensive to establish.
That has consequences across private markets.
Managers may need stronger operational infrastructure. Banks may need richer behavioural data. Administrators may become more important authentication nodes. Wealthy families may need to treat digital identity as seriously as legal ownership. Investors may devote greater attention to operational resilience alongside performance.
The result could be an emerging security premium for trusted intermediaries.
Private capital has historically rewarded relationships, confidentiality and specialist access. Those qualities will remain valuable. What changes is the evidentiary standard surrounding them.
A trusted relationship can no longer mean that a voice sounds familiar.
A familiar face cannot independently authorise a transfer.
A convincing email cannot prove that banking instructions are genuine.
The private-market institutions best adapted to the next phase may therefore be those that preserve the advantages of relationship-based finance while separating familiarity from authentication.
For UHNW investors and family offices, that is ultimately a wealth-preservation issue rather than a technology story.
Artificial intelligence does not have to defeat markets to destroy capital. It only has to persuade one trusted person to send money to the wrong place.
The institutional response is not fear. It is better architecture.
And in a financial system where identity itself can be synthesised, the scarce asset may increasingly be verified trust.
Works Cited
- https://www.ftc.gov/news-events/news/press-releases/2025/03/new-ftc-data-show-big-jump-reported-losses-fraud-125-billion-2024
- https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf
- https://www.cifas.org.uk/newsroom/fraudscape-2026-6-month-update
- https://www.theia.org/news/press-releases/cloning-scams-top-fraud-concern-investment-managers-and-their-customers
- https://www.ukfinance.org.uk/system/files/2026-06/UK%20Finance%20Fraud%20Report%202026.pdf
- https://www.interpol.int/News-and-Events/News/2024/INTERPOL-Financial-Fraud-assessment-A-global-threat-boosted-by-technology
- https://www.acaglobal.com/industry-insights/private-equity-firms-report-receiving-phishing-emails-and-scam-capital-calls/
- https://www.inc.com/amaya-nichole/hackers-targeted-major-wall-street-money-managers-cloned-voices-hedge-fund-stopped-them/91385954
- https://tollbit.forbes.com/sites/jessedamiani/2019/09/03/a-voice-deepfake-was-used-to-scam-a-ceo-out-of-243000/