A realistic face, voice or identity document can be generated or altered. Remote identity checking therefore cannot treat a convincing image or video as proof by itself; banks combine independent evidence, capture safeguards, behavioral signals and controlled escalation.
Deepfakes change the quality of the evidence
Deepfake tools can create or manipulate images, video, audio and documents used during account opening, authentication or servicing. The fraud may use an entirely synthetic identity or impersonate a real person whose data or credentials have already been compromised.
Visual realism is not the same as verified identity. A bank frames the decision around whether the claimed identity, presented evidence, live applicant and requested action are consistent, rather than asking only whether a selfie resembles a document photo.
Independent evidence reduces single-point reliance
Identity proofing can validate document and account information against authoritative or reliable sources and compare core attributes for consistency. Device, network, contact, application and prior-relationship signals can add context without turning any one data point into automatic approval or denial.
The layers should be appropriate to the action's risk. Opening an account, changing a phone number and sending a high-value payment can require different evidence and assurance even when the same customer interface is used.
Live capture and presentation-attack controls add friction for fakes
Live document capture and presentation-attack detection can make it harder to submit a replay, screen image, mask or injected synthetic stream. Active or passive liveness methods evaluate whether the sample appears to come from a present person interacting through the expected capture process.
These controls are probabilistic, not infallible. They can miss sophisticated attacks or reject legitimate users because of devices, disabilities, lighting or network conditions, so the bank measures performance and offers a secure alternate path rather than assuming every failed check is fraud.
Inconsistencies trigger step-up review
A mismatch among the face, document, device, contact details and expected behavior can move the request to stronger authentication or trained manual review. Depending on the risk, the bank may use a trusted callback, an existing authenticated channel, additional documentary evidence or an in-person option.
Reviewers receive clear indicators and decision criteria instead of an unexplained model score. Procedures address social engineering as well as media manipulation because a convincing synthetic applicant may pressure staff to bypass an ordinary control.
Monitoring connects proofing with later activity
Banks review confirmed fraud, false rejections, exception decisions and downstream account behavior to find weaknesses in the identity process. Testing includes new attack methods, vendor changes and the interfaces through which images or video could be injected.
Biometric and identity data are sensitive and require purpose, access, retention and security controls. Deepfake detection complements identity proofing, authentication and transaction monitoring; it does not replace them or guarantee that an apparently genuine applicant will use the account lawfully.
Read the primary material
Banking Explained prioritizes regulators, official publications and first-party announcements.