Your data. Your choice.

If you select «Essential cookies only», we’ll use cookies and similar technologies to collect information about your device and how you use our website. We need this information to allow you to log in securely and use basic functions such as the shopping cart.

By accepting all cookies, you’re allowing us to use this data to show you personalised offers, improve our website, and display targeted adverts on our website and on other websites or apps. Some data may also be shared with third parties and advertising partners as part of this process.

Legal noticePrivacy notice
Wikimedia Commons
News + Trends
95

ETH develops chip against deepfakes

Samuel Buchmann
25/3/2026
Translation: machine translated

A project at ETH Zurich is adding a cryptographic signature to photos and videos at hardware level. This should later be able to be read in a decentralised manner. This would allow platforms to automatically verify genuine recordings.

Fake videos and images of politicians, celebrities or the Pope: deepfakes are no longer a technical gimmick, but a business model for fraud and a risk to democracy. Misinformation and disinformation are on the rise, while countermeasures can barely keep up. As a result, many people no longer trust even credible sources.

Researchers at ETH Zurich want to counteract this problem with a hardware solution. They have developed an image sensor that cryptographically signs its data directly in the chip, making subsequent manipulations visible. Genuine images can be verified by anyone and everyone.

Decentralised verification via blockchain

The prototype not only generates the image data for a video recording, but also a digital signature. This is permanently linked to the respective sensor. It documents which chip the data originates from and when it was created. Any subsequent changes to the image material would invalidate the signature. In order to create a deepfake that passes as «genuine», attackers would have to compromise the hardware itself - i.e. physically attack the chip. This effort would be so high that counterfeiting would become economically unattractive for social media.

How the technology works: A real event (1) is recorded by a camera, whose sensor chip generates both the image data and a cryptographic signature (2). After storage in a public register (3), the signature can be used to check whether the recording is authentic and has not been altered (4)
How the technology works: A real event (1) is recorded by a camera, whose sensor chip generates both the image data and a cryptographic signature (2). After storage in a public register (3), the signature can be used to check whether the recording is authentic and has not been altered (4)
Source: Felix Franke / ETH Zürich

Camera manufacturers could store the signatures generated by the chip in a public register on a blockchain basis. Platforms or verification tools compare the signature of the file with the register during upload or verification. If it matches, the content is considered an authentic recording. If the entry is missing or does not match, this is a warning signal. The chain of trust should therefore shift away from platforms and software towards hardware.

  • Background information

    Why automated AI detectors don’t work

    by David Lee

Current detection methods usually look for fakes based on traces of AI generation in the image - such as unnatural artefacts or inconsistent lighting conditions. In contrast, the ETH approach does not work with probabilities, but with reliable proof of origin. For platforms such as YouTube or social networks, which are already testing deepfake reporting systems for politicians and journalists, the chip could automatically verify content.

For platforms such as YouTube or social networks, which are already testing deepfake reporting systems for politicians and journalists, the chip could automatically verify content.

Deeper integration than previous approaches

The sensor is still a prototype. The concept works for any type of sensor, from cameras to smartphones. For widespread use, manufacturers would have to adapt their devices and agree on common register standards. Whether this is realistic remains to be seen. Manufacturers such as Sony, Canon, Adobe and Microsoft have so far taken a software-based approach to verifying the authenticity of photos: C2PA is an open standard for the cryptographic signature of metadata. Some cameras already support it.

The prototype demonstrates the technical feasibility.
The prototype demonstrates the technical feasibility.
Source: Caroline Arndt Foppa / ETH Zürich

The ETH solution goes deeper. It transfers the generation of the signature to the chip and thus links the real event with the digital file at hardware level. Raw data and signature are inextricably linked. The second difference lies in the idea of the decentralised register. This means that it does not matter whether a company involved in the processing or transmission of the data is trustworthy.

Header image: Wikimedia Commons

9 people like this article

User Avatar
User Avatar
Samuel Buchmann
Senior Editor
Samuel.Buchmann@digitecgalaxus.ch

My fingerprint often changes so drastically that my MacBook doesn't recognise it anymore. The reason? If I'm not clinging to a monitor or camera, I'm probably clinging to a rockface by the tips of my fingers.

News + Trends

From the latest iPhone to the return of 80s fashion. The editorial team will help you make sense of it all.

Show all

5 comments

Avatar
later