Sony Uses Autofocus Depth Data to Detect AI Deepfakes in Video

Key Takeaways

• Sony’s Camera Verify software now supports video, using 3D depth data from the autofocus system to prove footage was shot with a real camera.
• The system analyzes three-dimensional depth information embedded in metadata — something AI video generators cannot replicate — to detect deepfakes and re-filmed screen captures.
• Camera Verify is currently in beta for news organizations and supports select Alpha, Cinema Line, and XDCAM cameras, with more models planned.

Sony has expanded its Camera Authenticity Solution to support video for the first time, giving newsrooms a way to verify that footage was actually captured by a camera — not generated by AI.

The key technical innovation: Sony’s Camera Verify beta software now analyzes 3D depth data from the camera’s autofocus system to confirm that a real three-dimensional scene was photographed. It’s a verification layer that current AI video generators simply cannot fake.

How Sony Uses Autofocus Depth Data to Fight Fakes

Camera Verify relies on three verification factors to authenticate content:

• C2PA digital signatures — A cryptographic signature is created inside the camera hardware at the moment of capture and embedded in the file in real time. The signing keys are stored in a secure hardware chipset, making them extremely difficult to forge.
• Tamper-proof timestamps — The capture time is acquired from a secure server and attached to the file. It cannot be altered after the fact.
• 3D depth detection — This is the breakthrough. Sony’s cameras record three-dimensional depth information alongside the image data, captured through the same optical axis as the photo or video. The software checks this depth map to determine whether the scene contains actual 3D subjects or is just a flat 2D surface — like a screen or a printed image being re-photographed.

That third factor is what makes Camera Verify genuinely different from other authentication approaches. Without depth verification, someone could theoretically photograph an AI-generated image displayed on a high-resolution monitor and still receive a valid C2PA camera stamp. The 3D depth map catches that trick — a flat screen reads as a 2D surface, not a three-dimensional real-world scene.

Why This Matters for the AI Authenticity Debate

Most current approaches to combating AI-generated imagery rely on watermarking, metadata standards, or AI-powered detection tools that try to identify synthetic patterns. All of these have weaknesses: watermarks can be stripped, metadata can be forged, and AI detectors produce false positives.

Sony’s approach is fundamentally different because it uses physics-based evidence. The 3D depth data comes from the camera’s autofocus system — the same system that uses phase-detection pixels on the sensor to measure distance to subjects. This information is captured on the sensor itself, along the same single optical axis as the image, which Sony says provides “information of high authenticity.”

Current AI video generators like Sora, Kling, and Veo produce remarkably realistic footage, but they work by predicting pixel values — they don’t have access to real-world depth data from a physical lens and sensor. Even as these models improve visually, they can’t produce the kind of autofocus-derived 3D depth map that a real camera sensor records.

Video Support: What’s New

Camera Verify originally launched in beta for still images. The expansion to video is significant because deepfake concerns are arguably even greater for moving footage — video carries more persuasive weight and is harder for humans to analyze frame by frame.

There are some technical constraints. Video verification only works with MP4 files recorded at a bitrate of 222 Mbps or lower. Footage shot in 8K or in slow/quick motion modes is not currently supported.

News organizations can validate both stills and video through a dedicated Camera Verify URL. Sony also offers an Image Verification Pre-Check SDK that organizations can integrate into their own editorial systems, allowing them to check for camera signatures without changing existing workflows.

Which Cameras Are Supported

Video authenticity recording is currently supported on the following Sony cameras:

Interchangeable-lens cameras: Alpha 1 II, Alpha 1, Alpha 9 III, Alpha 7R V, Alpha 7S III, Alpha 7 IV. Support for the Alpha 7 V is expected from May 2026.

Cinema Line: FX3, FX30

Professional broadcast: PXW-Z300

Sony has said additional camera compatibility is planned for the future.

Limitations and the Road Ahead

Camera Verify is still in beta and currently requires a paid license. It’s only available to select news organizations and photojournalists — it’s not a consumer feature. Sony notes that “the functionality and names of the features may change when the service is officially released,” and the official release itself isn’t guaranteed.

There’s also the obvious limitation that this only works for Sony cameras. While Sony is active on the C2PA steering committee helping develop industry-wide standards, the 3D depth verification is proprietary — other camera manufacturers would need to develop their own implementations.

Still, it’s the first approach we’ve seen that uses hardware-level physics data to verify content authenticity, rather than relying solely on software-based signatures or AI detection. In a landscape where AI-generated content is growing exponentially, that’s a meaningful step forward — especially for photojournalism, where the credibility of visual evidence is everything.

Frequently Asked Questions

Featured image: Photo by M Sohaib on Unsplash.