Cryptography to tackle deepfake, proving the photo is original

Question

Deepfake technology has become very difficult to tackle due to sophisticated machine learning algorithms. Now, even when a journalist or bystander provides photo or video evidence, the culprit denies it, claiming that it is the result of deepfake manipulation. Can TEE (Trusted Execution Environment) cryptography technology, like SGX, be used to validate whether a photo is original, taken directly from a camera, and free from any manipulation? This would ensure that the culprit cannot deny the authenticity of the photo. Does it require separate camera hardware, or can the right piece of software alone accomplish this? We can provide these special tools for journalists, etc., to decrease the harm caused by deepfake.

Answer 1

The best standard cryptography has to offer is that a photo can be signed by whoever attests that it is original. If we trust the signer and their ability to use their private key wisely (that is not let others use that key, and use it only to sign what they attest is original), and we trust the certification authority (even though some are under control of government actors), then anyone can verify that the photo has been signed and is unaltered, and existed at a certain date. There are issues with resize, crops, edit, but they can be solved by resigning, by the original signer or a trusted third party.

This is feasible with well-established cryptographic techniques, and useful. I hear of an emerging standard.

---

If we don't thrust the signer and want to put trust in the hardware, e.g. camera, then that's no longer a purely cryptographic problem, and it gets hairy, at multiple levels.

• For a start, what's "original"? Is anything imaged by a physical sensor "original"? What if it is imaged a high-resolution projection onto the sensor? What about physical reality, only staged?
• Trust in the camera signing subsystem. Trusted Execution Environments only make hacking harder, not impossible.
• It seems extremely difficult to guard against insider threats.

For these reasons, I doubt of the economic rationale of trusted hardware to certify photos, except perhaps in some rare contexts where the person that wants to be convinced of the reality of something prepares the hardware. This could apply in some situations: hostages, WMD inspections. Even then, good luck to convince whoever grants access to take the photos that the camera does not contain a GPS tracker or/and a microphone surreptitiously exfiltrating it's output in the certified photo? Or a plain explosive device?

Update: The above forgets that not-so-useful technology can nevertheless become reality in some segment of the market. It's in one recent, expensive camera.