Apps · Digital wellbeing and privacy · Web and API
Vera: a system that verifies claims, news and media
Is a claim going around about someone true? Was the video you watched or the voice you heard generated by AI? Vera runs text, audio and video through independent analysis engines and merges them into one report. No engine issues a final verdict: every report carries a probabilistic score, evidence and warnings. It is being built for journalists, agencies and anyone protecting their own reputation.
- In development
- Source credibility score
- Deepfake and voice-clone detection
- C2PA provenance check
- Evidence-backed probability, not a verdict
Why Vera?
Four independent engines
Audio, claim, video and fusion engines run separately; if one fails, the others can still deliver the report.
Catches contradictions
The fusion engine does not just average; it flags contradictions such as "the text looks true but the audio is AI-generated".
Careful language
Reports come with a probability score, an evidence list and warnings; never "definitely fake" or "definitely real".
One entry point
All engines sit behind a single gateway; text, audio and video are analysed in the same case.
What it does
Claim and source verification
Breaks a claim down with AI, cross-checks sources with Brave Search and returns a 0–100 source credibility score and an evidence list.
Audio deepfake and voice-clone detection
Estimates the probability that an audio file is AI-generated.
Video and face deepfake detection
Frame and face analysis produces a 0–100 image authenticity score and a list of suspicious frames.
Unified report (fusion)
Merges audio, claim and video results into one report and handles missing or failed engines gracefully.
C2PA content credentials check
Verifies a media file’s cryptographic provenance chain (C2PA Content Credentials).
Misinformation family tree
Maps where a claim came from, how it spread and how it mutated along the way, based on text similarity.
Trust Network
Matches new cases against a hash database of known deepfakes and misinformation samples.
Call protection
Aims to analyse audio in chunks during a live call and raise an instant voice-clone warning.
Telegram and WhatsApp bot
Send text or audio and get the report back in the chat.
Status
Today
- The four analysis engines and the gateway are built
- Not live as a public product yet
Goal
- A free top tier for consumers
- A team and subscription tier for journalists and agencies
Honesty and limits
Verification tools can create false confidence. Vera is being designed with that in mind.
- No engine issues a final verdict; results are probabilistic and do not replace human review.
- Every report carries evidence and warnings along with the score.
- Claim analyses about a person are worded with legal caution.
Important
Vera is a helper, not a decision-maker. AI detection results can be wrong; check sources independently before publishing or making legal or personal decisions.
Frequently asked questions
What is Vera?
A verification system that analyses claims, news and media about a person, assesses source credibility and tries to detect AI-generated media (deepfakes, voice clones).
Can it say for certain that a video is fake?
No. Vera gives a probabilistic score, evidence and warnings and issues no final verdict. Results do not replace human review.
What does it use?
An audio spoofing classifier for sound, a language model and Brave Search for claims, and face detection with an EfficientNet-based deepfake classifier for video.
Is it available to use?
Not yet. The engines and the gateway are built but it has not launched as a public product. Write via the contact page for early access.
Who is it for?
Journalists, agencies and anyone who wants to check content circulating about themselves.