Deepfake Endorsement

This week in one sentence: In the two cases that went on record this week, fraudsters copied not the offer itself but the trust signal that carries it. Evidence maturity: EARLY SIGNAL.

We look at three questions this week. What was actually copied beneath the headlines? Why does it matter now? Who pays the cost, and who will have to?

IA familiar face, a fake offer

The UK's fraud reporting service, Report Fraud, published an alert on 27 October. According to its data, £24.3 million was stolen through fake investment schemes between January and August 2026. In these schemes, AI and deepfakes were used to make the offer look more believable.

This is not the UK's total fraud loss. It is only the eight-month loss reported to Report Fraud for this type of scheme. Report Fraud received more than 1,300 reports of this type over the period. People aged 60 or over made nearly 80 per cent of them. The alert does not give the loss per person.

The mechanism is simple. Instead of assessing the offer, the person trusts the face presenting it. When AI can produce that face, trust itself has been imitated. These last two sentences are a CalibrateAtlas interpretation.

IIA fake filing, real money

The second imitation is not a face but a document. On 29 October the US Securities and Exchange Commission (SEC) charged two companies. They are BG Wealth Sharing Ltd and DSJ Exchange Pty Ltd. According to the SEC, the companies used false SEC filings to build confidence in their investment programme.

The SEC alleges that this method raised more than $130 million from tens of thousands of retail investors. This is a charge. It is not a court ruling. The figure is also the SEC's allegation.

What the two cases share is this: both imitate an external endorsement that an individual cannot easily verify on their own. In one it is a familiar face, in the other a filing that appears to have been made to the regulator. This comparison is a CalibrateAtlas interpretation.

To use an analogy, both are like faking the hygiene certificate in a restaurant window. Customers do not walk into the kitchen to inspect it. They look at the certificate in the window and go in. If the certificate is fake, the customer's shortcut has been turned against them.

IIIHow is a genuine signal registered?

If the fake can be produced, how is a genuine trust signal proven? In the same week, several institutions on the digital asset side went through official channels.

Ripple announced that the OCC, which supervises banks in the US, had granted final approval for Ripple National Trust Bank. This rests on Ripple's own announcement.

Galaxy applied to the same agency to establish a national trust bank. An application, not an approval.

BNY said it now supports XRP, the native digital asset of the XRP Ledger, and RLUSD on its digital asset custody platform. DTCC's tokenization service went live. The service is currently connected to the Canton Network and DTCC's private Besu network. More networks are due in 2027.

None of these announcements disclosed usage volumes. A licence and a live service change who supervises the asset. They do not tell you how many people use it.

Not everything in this layer is fast either. A Mastercard executive warned this week against a "stablecoin sandwich". He meant a fast transaction in the middle, with slow entry into and exit from the stablecoin. C2FO and Ripple, for their part, announced an initiative to move financing and settlement activity onto the XRP Ledger. That too is a plan. It is not yet a running operation.

IVThe bill for control

A lack of control has its cost too. Several examples of that cost went on record this week.

OFAC, which enforces sanctions in the US, announced a $1,529,685 settlement with the networking technology company Extreme Networks. Extreme Networks is not a bank.

Malaysia's central bank, Bank Negara Malaysia, penalised AmBank and AmBank Islamic. Each bank was fined RM276,000. The reason was failing to submit cash threshold reports on time. The penalty was imposed on 10 September and announced on 28 October.

Expectations are rising on the security side as well. In an a16z survey of 100 chief information security officers, 87% expect cybersecurity spending to increase. Their reasons are AI-driven threats. The attack surface widened by their own use of AI is one of them. The survey is an investment firm's publication and rests on what respondents said.

An example of what defence might look like came from the UK. In its review of fair value in consumer finance, the Financial Conduct Authority (FCA) cited one firm's practice as good practice. That firm used comprehension testing to check whether customers understood when charges applied. This is not a rule. It is a good-practice example the FCA described.

Regulators' interest in this area was visible before this week. On 23 October the FCA restricted six lenders over gaps in their anti-money-laundering controls. On 7 October the Monetary Authority of Singapore (MAS) issued AI risk management guidelines for financial institutions. Both are records from earlier weeks, not this one.

VWhat the evidence says, and what is missing

First, the limit. The records we have contain no comparison with earlier periods. So we cannot say "this type of fraud has grown by this much".

But the significance lies in the mechanism, not the size. A familiar face's endorsement is one of the signals individuals trust most. That signal can now be produced with technology. Report Fraud set out this type in a separate alert with its own figure. This new area of fraud, opened up by technology, has become visible in the official record. This is a CalibrateAtlas interpretation. The record shows that this type exists. It does not show how large it will become.

The same caution applies to institutions' own figures. This week Visa said its stablecoin settlement business is running at an annualized rate of more than $25 billion. This is a company statement. It is not an independent measurement.

For CalibrateAtlas's payments infrastructure thesis, the picture did not change this week. DTCC's service went live and BNY expanded custody. Neither disclosed volume. The usage data the thesis is looking for still exists only in institutions' own statements.

VIWhat falls to individuals and institutions?

For individuals, what changes is less the figure than the signal itself. A familiar face's endorsement is ceasing to be proof of trust on its own. So is the phrase "officially registered". No data this week shows how often individuals verify these signals.

For institutions, the question is different. How will a customer verify that a signal in front of them really belongs to the institution? In the sources we reviewed this week, there was no measured answer to that question. The idea that this burden is shifting to institutions is a CalibrateAtlas interpretation.

VIIWhat to watch

Let us return to the face from the opening and the certificate in the window. Individuals struggle to tell on their own whether either is genuine. So the burden of verification shifts to institutions and regulators. This too is an interpretation.

Four signals could change this picture in the coming weeks. Will Report Fraud or another official source publish data comparing this type with earlier periods? How will the SEC's case against BG Wealth and DSJ end? Will DTCC or BNY disclose real usage data? Will Galaxy's application be approved?

Correction, 2 November 2026: The first version of this article said the Report Fraud alert did not give the number of victims. The alert reports more than 1,300 reports of this type, with people aged 60 or over making nearly 80 per cent of them. The sentence has been updated. A report count is not the same as a victim count.