AI Composites Passed Hong Kong Bank Identity Checks Thirty Times and Funded a Fraud Network

What happened

Eight people were arrested in Hong Kong in April 2025 after police connected them to a scheme that used AI-generated facial composites to pass bank identity checks. The group submitted 44 account applications using ID photographs that had been altered with synthesized faces. Thirty of those applications succeeded. The checks designed to verify that an applicant is who they claim to be processed the images, returned approvals, and opened the accounts.

The core of the scheme was a gap in how online identity verification reads a face. Most systems used by financial institutions check that a submitted photo contains a face consistent with the ID document and sometimes confirm that a live selfie matches it. They are built to catch lazy fraud, stolen credentials, and low-quality fakes. AI-generated composites that are photorealistic and structurally consistent with a document image can sit above the threshold those systems were calibrated against. In this case, they did, in six out of every seven attempts.

Once the accounts were open, they functioned exactly as legitimate accounts do. The fraud ring used them to apply for loans and run credit card purchases totaling HK$860,000, then channeled more than HK$1.2 million in suspected criminal proceeds through the same accounts. Police linked the network to local triad-affiliated syndicates, suggesting the operation was structured rather than opportunistic and that the technique was in active use rather than being tested.

The 68 percent success rate is the figure that matters most. It does not describe a sophisticated attack against a weak system. It describes a repeatable technique against systems that are operating exactly as designed, just not designed for this. Identity verification that works against stolen documents and low-resolution forgeries has a different failure boundary than identity verification that works against a synthetic image trained on a large corpus of real faces. That boundary was never moved, and this scheme found it without effort.

What the record does not contain is an audit trail for the checks themselves: which applications a given system processed, what confidence score each returned, and whether those scores clustered in a band that should have flagged the batch as anomalous. Thirty approvals from the same fraud network is a pattern, and patterns are only detectable if the checks produce a provable record of what a system did and when. Without that record, identity verification is a checkpoint with no memory of who passed through it.

Reported impact

Affected parties

Not publicly disclosed

Harm type

Not publicly disclosed

Scale

Not publicly disclosed

Financial impact

Not publicly disclosed

Regulatory action

Not publicly disclosed

Classification

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.