Earlier this year, the US Government Accountability Office (GAO) was warning that deepfakes – ie. video, audio and image content that seem real but are actually manipulated with AI – had been used to try to influence election results.
The GAO also advised that, in response to President Biden’s 2023 executive order calling for private industry to step up, companies were now developing improved methods to detect deepfakes, most notably by labelling AI-generated content with secure watermarking.
A leading company in this space is San Diego-based Verance, whose digital watermarking technology is recognised as the industry standard for movies, music and TV. In advance of the 2024 US elections, the company launched Verance Authentication, for ensuring the integrity and authenticity of digital content, particularly in combating AIgenerated disinformation.
Nil Shah, CEO of Verance explained in an article for Variety.com how the technology works in partnership with cryptographic signatures, while emphasising that cryptography alone is not enough to ensure authenticity.
‘The power of cryptography is it will tell you with absolute certainty when even a single byte of the digital media has been changed.
What it can’t do is tell you whether the content of the digital media has indeed been altered,’ he said.
‘This is a considerable limitation, as the bytes in digital media change through routine handling in nearly every step of distribution — be it broadcast, streaming or social media — even when the actual content isn’t modified at all. So a solely cryptographic tool can’t be practically used to authenticate media distributed in the normal ways people get their media today.
‘That’s where digital watermarking comes in,’ explained Nil Sha. ’These invisible labels can be embedded in content before it is distributed, remain readable even after the content has been distributed and carry information (for example, a URL) linking that content back to its original, cryptographically protected form.
‘Any distribution platform where the content appears downstream, such as a social media site, can read the watermark. And if the content cannot be authenticated immediately, one can follow a link to access the original, authentic version that has its cryptographic signatures intact.’
He added that anyone watching the watermarked content on their TV, computer or mobile device, would see a small icon identifying the content as trusted.
Furthermore, the viewer would be able to click on the icon and link back to the original transcript or thumbnail image of the content.
