(Credit: René Ramos)
LAS VEGAS—Like it or not, generative AI is a part of our online and offline worlds now. It's no wonder that discussions about detecting deepfakes or other digitally altered or created content were hot topics of interest at this year's Black Hat cybersecurity conference.
One of the last briefings at the show was from Adobe Principal Scientist Peleus Uhley, who explained Content Credentials. Think of them as nutrition labels for digital media (like ISPs are doing for broadband service). They reveal valuable cryptographic data that verifies the validity of an image and confirms when and how the content was created. The label tracks modifications made over time, and people can use the label to request that the media isn't used to train AI models.
(Credit: C2PA)A little "CR" bubble logo on digital media identifies the credential. Click or tap the label, and it will show you all of the information it contains.
You can already see the labels in action on newer Leica cameras, such as the M11-P. You can also enable them while posting on LinkedIn or when creating digital media using Microsoft's software.
Math Doesn't Lie, People Do
The Content Credential labels are provided by the Coalition for Content Provenance and Authenticity (C2PA), which consists of hardware, software, photography and online digital media companies.
At Black Hat, Uhley stressed that the content labels are not designed to detect deepfakes, and they aren't instant indicators of factual evidence. A person or a news organization can still spread lies while using images affixed with a "CR" label. The labels are for news organizations and others to claim their content and show if or how that content has been modified over time.
Before attending Black Hat, I spoke with Andy Parsons, who is the senior director of Adobe's Content Authenticity Initiative. He told me the cryptographic labels are necessary because "math doesn't lie, people do." C2PA members hope that the labels will inspire public confidence in creating and sharing digital media while also combatting the spread of misinformation or disinformation via digitally altered images or deepfakes.
The labels aren't available on every online platform yet, and Parsons likened the markers' rollout and adoption to passkey technology. When passkeys were first introduced a few years ago, the public response was fairly muted because it took time for companies to adopt the technology. Today, widespread passkey adoption is ramping up, and you can use passkeys to log into many of your online accounts, including those for Amazon, Google, and Microsoft.
An Images' History at Your Fingertips
Parsons says that widespread adoption is the long-term goal for these labels. He suggested they could become as ubiquitous as the copyright symbol, and could potentially help people have more confidence when interacting with online media in the future. Ultimately, the idea behind the labels is that "people have a right to know how things work," he said, and the public deserves transparency about the creation and history of the media they're consuming.
Parsons is also realistic about the labels' role in the fight against misinformation and deepfake technology online. He told me that the best method for deepfake detection right now is to use your eyes, ears, and intuition when consuming any online content. Generative AI tools are getting better every day, so if you see an image or video that seems too good or too terrible to be true, take a moment and pause to do some fact-checking before sharing the content.
In a similar vein, my PCMag colleague Chandra Steele recently put together a primer for spotting AI-generated content online, complete with a quiz to test your AI-spotting skills. Check it out, and tell us your score in the comments.