Can You Trust Video Evidence?
Convincing synthetic audio and video are emerging, but courts are not yet flooded with deepfake challenges. How to preserve originals and assess contested recordings.
Short answer
Deepfakes create two risks: fabricated recordings and genuine recordings dismissed as fakes. Most clips are still authentic, but high-stakes evidence needs the original file, a chain of custody, corroborating records, and forensic advice when authenticity is disputed — not a single detector score.
A recording used to be the thing that settled an argument. If two witnesses disagreed about what somebody said, people wanted to hear the audio or see the video. AI is making that instinct less reliable.
On 15 October 2026, representatives of OpenAI and Meta warned a United States judicial evidence panel that convincing false audio and video can be created quickly and cheaply. Reuters reported that the panel established a group to examine the problem. The discussion was not a finding that courts are already overwhelmed by fake recordings. In a survey cited at the meeting, only 15 of 931 judges reported cases involving challenges to alleged deepfake evidence.
That distinction matters. The danger is emerging, but the response should be careful rather than panicked. Most recordings do not suddenly become worthless because convincing fakes exist. Equally, an apparently realistic clip should not be treated as unquestionably genuine.
There are two risks. The obvious one is that someone manufactures a recording showing words or actions that never occurred. The less obvious one is that a genuine recording is dismissed as a fake. Either can harm an employee, a customer, a litigant, or a business.
Imagine a voice recording of an executive authorising a payment, a video submitted with an insurance claim, or an interview used in a workplace investigation. Once the original file has been lost and only a forwarded clip remains, it may be difficult to establish who created it, whether it was edited, and where it has been stored.
Software that claims to detect deepfakes may help an investigation, but its score should not settle the question by itself. An authentic recording can be compressed or altered by an ordinary application, and a fabricated recording may be designed to defeat detection. Sound evidence requires more than a single technical test.
What to do
Preserve the original recording wherever possible. Keep the full file rather than just a short extract, along with its creation details and the record of how it was collected. Record who received it, when copies were made, and whether any editing occurred.
Ask whether independent material supports the account: other recordings, messages, calendar entries, witness accounts, or transaction records. Where authenticity is contested or the stakes are high, obtain suitable forensic advice before drawing conclusions. Businesses should also have a process for responding fairly when someone claims that genuine material is fabricated.
Sources
- Reuters, 15 October 2026 — OpenAI and Meta testimony on deepfake evidence.
- ISO/IEC 42001 explained — International Organization for Standardization.
We help legal, HR, and investigation teams define retention, verification steps, and ownership of disputed AI-affected evidence — as part of gap assessment and ongoing AIMS support.