Verify whether the media driving an attack is real
In short. Signal by AI Uniti scans the audio, video and images in a campaign for AI manipulation, and scores authenticity alongside the coordination signals around your brand. You see whether the media is real and who is amplifying it, in one platform.
Coordinated campaigns increasingly carry a manufactured artefact at their centre: a fabricated image, a cloned voice, a doctored clip. Detecting the media is one half of the problem. Detecting the coordinated network manufacturing its reach is the other. Signal does both, in one platform.
The authenticity gap
AI-generated media is now, in many cases, indistinguishable from real media to the human eye and ear. The old advice, look for the visual glitch, no longer holds, because the glitches have largely gone.
Provenance signals help only when the generator cooperates. Content credentials such as C2PA and watermarking approaches can mark media as AI-generated at the point of creation, but most malicious media carries no provenance: it is stripped, re-encoded, or produced by a model that never added it. So detection cannot depend on metadata. It has to read the media itself.
The cost is already landing on businesses:
- US$23 billion - expected annual cost of synthetic fraud by 2030 (Deloitte).
- 92% - of businesses have incurred deepfake-related financial losses (Regula).
- 704% - increase in deepfake face-swap attacks on identity-verification systems (SC Media).
- 57% - of attacks involved AI-generated document forgeries in 2024, up from close to zero in 2021 (LexisNexis).
What your existing stack misses
Social listening and monitoring tools were built to measure volume and sentiment: how much is being said about you, and whether the tone is positive or negative. They can tell you a clip is spreading. They cannot tell you whether the clip is real, and they cannot tell you whether the accounts pushing it are a coordinated network or a genuine audience. For brand, communications and risk teams, that is the gap that matters, because the damage is done by manufactured media amplified by a manufactured audience.
How Signal detects synthetic media
Signal surfaces authenticity through three honest surfaces:
- Inside brand monitoring. Media collected in Signal’s brand and executive monitoring is scanned for AI manipulation, and the authenticity verdict is surfaced next to the behavioural and coordination signals. You see whether the media in a campaign is real in the same place you see who is amplifying it.
- On-demand scanning. Paste a URL or upload a suspect image, clip or audio file, and get an explainable verdict in seconds through Defend, part of the Unite defence layer.
- Agent-native. An analyst can ask an AI assistant to check a piece of media, or summarise a brand’s authenticity, executed as a governed, identity-scoped and audit-logged tool call. More on the Unite defence layer.
We are extending detection into voice, meetings and identity
Impersonation increasingly targets the live channel. AI Uniti’s content-authenticity detection already scores AI-generated audio, video and imagery, and we are actively extending it into three high-risk areas:
- Voice fraud - cloned voices defeating contact-centre authentication.
- Deepfakes in meetings - AI-generated participants in high-stakes video calls.
- Identity and KYC - synthetic faces and documents defeating verification.
If any of these is a live risk for your organisation, talk to us.
Detection plus coordination
This is the difference. Authenticity of the media answers one question: is this real. Behavioural detection answers the other: who is pushing it, and are they acting together. A manipulated clip amplified by a coordinated inauthentic network is a larger, faster threat than the same clip sitting still, and the response is different too.
The same platform that detects coordinated narrative manipulation targeting your organisation now verifies whether the media driving those narratives is real. That combined view is one that neither a deepfake-only tool nor a social-listening tool offers. The behavioural side is powered by PulseCheck, which scores the accounts amplifying a narrative on a bot-to-human spectrum.
Model-agnostic and explainable
Detection is model-agnostic: it detects AI-generated content regardless of which generator produced it, and it does not depend on provenance metadata or watermarks. It returns a probability-based risk score, an ensemble manipulation probability mapped to authenticity bands and a severity score, with a per-signal breakdown, rather than a binary pass or fail. No biometric enrolment and no personal identity data are required. To understand the science, see how deepfake detection works.
Frequently Asked Questions
How does Signal detect deepfakes?
Signal scans the media captured in the campaigns it monitors, and lets you scan a suspect image, clip or audio file on demand. Detection models read the media itself for the statistical traces generative models leave behind, and return a probability-based authenticity score with a per-signal breakdown, not a binary pass or fail.
Can you detect synthetic media without a watermark?
Yes. Detection is model-agnostic and reads the media directly, so it does not depend on provenance metadata, C2PA or watermarks. That matters because most malicious media carries no provenance at all.
How is this different from social listening?
Social listening tracks mentions and sentiment. It can tell you a clip is spreading; it cannot tell you whether the clip is real, or whether the accounts pushing it are coordinated. Signal adds both: content authenticity and behavioural coordination, in one view.