Integrity atthe Core.
We sign audio and video from inside the signal. Anyone can check who published it, for free.
We usually reply within one business day.
The problem
A video call, a voice note or a video statement can be fake and look real. Detecting fakes is a race that is always one step behind.
was transferred by an employee of a Hong Kong firm after a video call with a fake chief financial officer (January 2024).
Source: Deloitte, 2024in fraud losses that generative AI could enable in the US alone by 2027, up from US$12.3B in 2023. A projection.
Source: Deloitte Center for Financial Services, 2024Since that day, Article 50 of the EU AI Act applies: providers of generative AI must mark the audio and video it produces.
Source: Regulation (EU) 2024/1689Regulation already asks for synthetic content to be marked. Authentic content must be provable too.
Why Kernel
Metadata gets lost on the first reshare. Our signature lives inside the audio and video itself, built to withstand what happens to them every day: compression, re-encoding and metadata stripping.
AI detectors guess what is fake. We prove who signed what is authentic, with digital-signature cryptography (Ed25519): a proof for each channel, not a probability.
Try it
Two sample videos: one signed by our team and one unsigned. Verify them for free, with no account, and compare the results.
Verify the samples
Signed in our console, just as any customer would. The result shows who signed it and their organization.
Result: Verified

A clip that has never been signed with Kernel.
Result: No Kernel mark detected
Comparison
Not everything called "authenticity" answers the same question.
Kernel and C2PA are complementary: Content Credentials tell the file's history, and Kernel's signature stays in the signal when that metadata is lost.
Who it's for
Sign your videos and audio when you publish them. When they travel beyond your site, your readers can check they're yours.
If a fake video of a spokesperson appears, anyone can check which one carries your signature and which doesn't.
Your voice and your image are your brand. Sign what you publish so your audience knows what's yours.
Build signing and verification into your product with our REST API.
The Origin
We were born in research: our founder has peer-reviewed scientific publications. If it can't be proven, we don't promise it.
Signal processing, linear algebra and modern cryptography. Math isn't decoration on our product: it is the product.
Every change goes through automated tests and review before it reaches production, and we measure robustness with test batteries on real audio and video. We care about the epsilon, not just the marketing.
Deterministic verdicts: same file, same result, every time. And we publish our limits instead of hiding them.
FAQ
You sign your audio or video from your browser and the signature stays inside the signal, travelling with the file as it is shared. Anyone can then check it for free on Verify: who signed it, proven channel by channel. See it in 30 seconds.
A European deep-tech startup based in Galicia, Spain, that signs audio and video from inside the signal. The signature is bound to its author's identity, so anyone can later check who published a file, with cryptographic proof.
No, and that's by design. Detectors use AI to guess whether something is fake, and they can get it wrong. We go to the source: authentic content is signed when it's published and anyone can check who signed it. We don't pass judgement on content that was never signed with Kernel.
Yes, in everyday use: it's built to withstand compression, re-encoding and metadata stripping. Heavy edits can remove it from one channel or both. The audio and the video are checked separately, so the result says which one still proves the signature (Verified, Attested or No match).
They're complementary. C2PA Content Credentials travel in the metadata, which many platforms strip when they re-encode. Kernel writes the author's signature into the signal itself, so it can still be read once the metadata is gone.
Article 50 of the EU AI Act requires AI-generated or manipulated content to be marked or disclosed. Kernel Security covers the other half: letting genuine content prove who published it. It strengthens AI Act transparency measures, though on its own it does not make an organisation compliant.
Anyone, for free and without an account: upload the file on the public verification page. Only signing is paid, with plans for creators, newsrooms and institutions.
Kernel Security is based in Galicia, Spain, and was founded by José Manuel Ramírez Sánchez, a researcher with peer-reviewed publications. It is qualified by the Xunta de Galicia as a Technology-Based Employment Initiative (IEBT) and was selected for StartTIC, the high-technology incubator of Zona Franca de Vigo.
We'll show you Kernel with your own files and work out together how it fits the way you publish.
We usually reply within one business day.