Recognizing the limits of deepfake detection
Understand synthetic voices, images and video, why visual clues aren't proof, and how to check the source before acting on or sharing business content.

A familiar face or voice can make a message feel credible. Synthetic media can imitate both, and an ordinary edited recording can also misrepresent what happened. Staff need a way to check content without having to become forensic analysts.
For a business, the important questions are where the material came from, what it claims and what someone wants you to do with it. You don’t need to prove that a recording is fake before pausing an unusual request.
Understand what can be imitated
Deepfakes use AI to generate or alter images, audio or video that imitate a person. A voice message might sound like a manager, while a video might place a familiar face in a conversation that never happened.
The Cyber Centre’s social-engineering guidance explains how synthetic media can support impersonation. It warns against treating audio or video alone as proof of identity.
AI-written email can imitate someone’s style too, but it isn’t the same technical category as a deepfake recording. The business response still needs to verify the request.
Treat visible flaws as clues
Odd lighting, unnatural pauses or a mismatch between lips and speech can prompt further checks. They can’t tell you on their own whether a recording is real. Compression, a weak connection and ordinary editing can create similar effects.
A recording with no visible flaws can still be fake. Don’t let the decision to act depend on staff spotting a distorted hand or an unusual accent.
If a detection service is proposed, ask what material it supports and how its results will be interpreted. Don’t upload private meeting recordings, client information or staff images to a detection website the business hasn’t checked, just for a quick verdict.
Trace the material to its source
Separate the person who sent you the clip from the person or organization it claims to show. A repost, screenshot or forwarded attachment can lose context and provenance.
Look for the original publication through an independently found official channel. Check the date, surrounding statement and whether the clip has been shortened. For internal content, contact the responsible person through your established directory or business system.
An authentic recording can still be used out of context. Verification should cover the claim as well as the identity.
Follow your usual sign-off, however real it looks
If content asks for money, account access or confidential records, follow the approval process you already use, however convincing the voice or live video is.
Use the request-verification guide for independent callbacks and escalation. The supplier-payment guide covers bank-detail changes and how payments get signed off.
Preserve evidence without spreading it
Report suspicious material through the agreed internal route. Keep the original message, link, time and requested action available to IT or the person handling the incident. Avoid circulating the clip as entertainment or publicly accusing the person it depicts.
If someone acted on the request, say what information, access or payment was involved. IT and the business owner can assess the next steps. Watching a manipulated recording doesn’t mean malware was installed.
Practise with fictional examples. The exercise should test whether staff can pause, find an independent source and report a concern, even when they can’t identify how the content was made.


