AI Aware
AI Aware is a British AI content detector covering four modalities in one platform: text, deepfake video, images and cloned audio. It returns calibrated probability scores paragraph by paragraph rather than a binary verdict, and starts free.
What is AI Aware?
AI Aware is an AI content detection platform built by Happy AI Ltd, a British deep tech company founded in 2023 and based in London. Its distinguishing claim is breadth: where most detectors handle a single content type, AI Aware covers four in one place, namely written text, video, still images and audio. That matters because a single fraud attempt now routinely arrives as a cloned voice note, a deepfake clip and an AI-written message at once.
Technically, the platform runs an ensemble framework rather than one classifier. Several detection models with different training data and different analytical approaches are combined, and machine learning is deliberately mixed with statistical, non-machine-learning techniques. The company argues this makes it more resilient against so-called out-of-distribution inputs, meaning content from generators the system has never seen, and against hybrid material where a human has edited AI output or an AI has altered human writing. Statistical fingerprinting looks for traces that survive paraphrasing: patterns in token probability distributions, characteristic phrase structures, and spectral signatures in audio and video.
Each modality has its own signals. Text analysis works at paragraph level and feeds a visualisation tool that shows whether AI content is concentrated in one block or scattered throughout. Video detection examines facial inconsistencies, unnatural blinking, lighting artefacts and temporal anomalies across frames. Image detection reads pixel-level noise, texture repetition and metadata inconsistencies, and flags hybrid manipulation such as inpainting or background replacement. Audio detection examines pitch variation, breath placement, micro-pauses, vocal resonance and background noise signatures, then transcribes the recording and analyses the wording as well.
Results come back within seconds as calibrated probabilities rather than a single percentage. On its own benchmark, derived from an extended version of the M4 dataset, the company reports 98 per cent accuracy, precision above 99 per cent and a document-level false positive rate below one in a thousand, after testing on more than five million pieces of real-world content. The research behind it comes partly from City St George's University of London, where co-founder Dr Tillman Weyde is a Reader in Computer Science, and development has been funded by Innovate UK grants. Access is through a web application, or through an API arranged on request.
What it does
- Detect text produced by large language models, including ChatGPT, Claude, Gemini, Grok and LLaMA
- Detect deepfake or AI-manipulated video from generators such as Synthesia, Sora, Runway, HeyGen and Seedance
- Detect AI-generated or AI-retouched images from Midjourney, DALL-E, Stable Diffusion, Firefly and Getty Generative AI
- Detect cloned voices and synthetic speech, including output from tools such as ElevenLabs
- Show where AI content sits inside a document, paragraph by paragraph, through the visualisation tool
- Return a calibrated probability score instead of a yes or no verdict
- Analyse the audio of live video and phone calls in real time, on request
When to use AI Aware / When not to
A quick filter to help you decide if AI Aware is the right fit.
When to use AI Aware
- Universities and schools running academic integrity checks on essays, dissertations and coursework
- Law firms, compliance teams and forensic investigators authenticating documents, recordings and submitted evidence
- Publishers, newsrooms and editorial teams verifying articles, images and footage before publication
- Recruiters and HR teams screening CVs, written tasks and remote video interviews
- Fraud, KYC and trust-and-safety teams in banking, insurance and online platforms
When not to use AI Aware
- Individuals who want to compare prices and buy in self-service, since no rate is published anywhere
- Teams that need a mobile app or a browser extension, as the product is web only
- Anyone looking to generate, rewrite or humanise content rather than analyse it
- Organisations that need a signed DPA, a subprocessor list or a documented hosting location before any upload
- Cases where a single automated verdict must stand as proof, which the vendor itself advises against
How to use AI Aware
A typical end-to-end flow, from setup to results.
- Open the detector directly from the site: no credit card and no sales call are required
- Run a one-off image check straight away, which needs no account at all
- Register an account to receive free credits and keep a history of your checks
- Upload the item you want assessed: a document, an image, an audio clip or a video file
- Wait a few seconds for the result, as there is no processing queue
- Read the calibrated probability score rather than treating it as a single overall percentage
- Open the visualisation tool to see which sections carry the AI signal and how they are distributed
- Buy credits when you need to analyse larger volumes of content
- Contact the team for bulk access, batch processing, API integration or bespoke reporting
- Ask the data science team for a written statistical report when the decision carries high stakes
Pros & Cons
Pros
- Four content types covered by one platform, which is uncommon in this market
- Explainable output: calibrated scores plus paragraph-level localisation, not a single opaque number
- Benchmark published openly, with named competitors, a stated dataset and a described method
- Genuine academic grounding, with a co-founder who is a Reader in Computer Science at City St George's
- Innovate UK grant funding, awarded to roughly five per cent of applicants according to the company
- Free entry point with no credit card, and one-off image checks that need no account
- Clear British jurisdiction, a named Data Protection Officer and no sale or rental of personal data
Cons
- No price is published anywhere: no pricing page, no amount, no tier, everything goes through sales
- No email address at all on the site, so every enquiry runs through a form with a two-working-day reply
- An API is advertised on three product pages but has no public documentation
- No mobile app and no browser extension, so the product is web only
- No security certification is displayed, neither SOC 2 nor ISO 27001 nor Cyber Essentials
- No data processing agreement, no subprocessor list, no stated retention period and no hosting country
- Performance figures are self-declared and rest on a benchmark the vendor ran on its own dataset
Pricing & Plans
There is a genuine free entry point: the image detector can be used immediately for individual checks without an account, and registering grants free credits, with no credit card required at any stage. Beyond that, AI Aware publishes no price at all. Additional volume is bought as credits, and bulk access, API integration or bespoke reporting are arranged as enterprise agreements through the contact form. No amount, currency or tier appears on the site or in the application, so the lowest paid price point cannot be stated.
- free
- no account required
- free credits granted on registration
- no credit card
- purchasable for analysing larger volumes
- price not published
- bulk access
- batch processing
- API integration or bespoke reporting
- quoted on request
Data, GDPR & hosting
A consolidated view of how AI Aware handles your data.
GDPR overview
The words GDPR and data protection regulation appear nowhere on the pages reviewed, and the company never claims compliance in so many words. What it does publish is closer to the substance: a named Data Protection Officer, Stephen Harmston, reachable at Exchange at Somerset House, London WC2R 1LA, though without an email address. The privacy policy, last updated on 6 May 2026, grants a right to export the personal data held and a right to have data and uploaded files erased, with two carve-outs: records kept for administrative, legal or security reasons, and data already folded into aggregated model training. Missing entirely are a data processing agreement, a subprocessor list, an Article 27 representative and any stated retention period. Governing law is England and Wales.
Who owns the data?
AI Aware is the trading name of Happy AI Ltd, registered with UK Companies House under number 14749356. Files you upload remain your content, but the terms of use let AI Aware review those submissions to monitor the accuracy of its detection systems, and the privacy policy states that uploaded content is collected and stored to deliver and improve the service, with the option to opt out. The company says it does not sell, trade or rent personal information; it may share aggregated information with service providers bound by confidentiality, and may disclose data where the law requires it. The site itself and its contents remain the property of the publisher and its licensors.
Reuse rights
Reuse of the site's own material is tightly limited: the terms forbid reproducing, distributing, modifying, republishing or transmitting content, with three exceptions only, namely temporary copies held in RAM while browsing, files cached automatically by the browser, and printing or downloading a single copy of a reasonable number of pages for personal, non-commercial use. Reverse engineering the detection techniques is expressly prohibited. Detection results are supplied as AI Analysis, and the terms place the responsibility for evaluating and verifying them entirely on the user before any reliance or action; no clause grants an explicit commercial licence over those outputs.
Data retention & training
Hosting summary
The site says nothing about where analysed content is processed or stored: no country, no region, no cloud provider is named on any of the pages reviewed, and no security certification such as SOC 2, ISO 27001 or Cyber Essentials is displayed. What is established is the legal frame rather than the technical one: the publisher is a UK company and the terms of use place any dispute under the law of England and Wales, before the courts of England and Wales. Network checks show the marketing site running on WordPress at IONOS SE, on an address geolocated to London, but that describes the brochure site, not the detection service, which runs on a separate subdomain whose hosting is undocumented. On security, the privacy policy commits to protective measures and to two-factor authentication where possible, while explicitly passing the residual risks of internet transmission back to the user.
Things to keep in mind
Risks and trade-offs to weigh before adopting AI Aware.
- A false positive on an essay or a job application has real consequences, and the vendor itself accepts that no detector is perfect
- The terms place responsibility for verifying every result on the user, so acting on a score alone shifts no liability back to the company
- Uploaded content is stored by default; the opt-out exists but has to be exercised rather than assumed
- Data already aggregated into model training is excluded from the right to erasure, so deletion is not always complete
- Sensitive material such as legal files, CVs and evidence would be uploaded to a third party with no published DPA or subprocessor list
- No hosting country is disclosed, so where the analysed content is processed cannot be verified
- A score is easy to substitute for human judgement, which is exactly the reflex this kind of tool should not encourage
Setup & Integrations
Technical difficulty
Almost nothing to set up. The detector is a web application: you open it, upload a file and read the result within seconds, with no installation, no configuration and no sales call. A single image check needs no account at all, and registering for free credits takes a couple of minutes. The only path that requires technical work is the API integration, and that one is harder than it looks, because there is no public documentation: scope, authentication and limits all have to be agreed with the team first.
Deployment
Behind AI Aware
Fundraising
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement AI Aware.
Frequently asked questions
What can AI Aware actually detect?
Can it catch text that has been rewritten to avoid detection?
How accurate is it, and how often does it flag human writing by mistake?
What does a calibrated confidence score mean here?
How much does it cost?
Is there an API?
Will my uploads be used to train the models?
Is there a minimum age, and who is behind the tool?
Should you pick AI Aware?
AI Aware occupies a narrow and defensible position: while most detectors examine one content type, it examines four, and it insists on explaining itself rather than issuing verdicts. The paragraph-level scoring and the visualisation tool are the substance of that claim, because they let a teacher, a solicitor or an editor see where the signal sits before deciding what it means. The academic grounding is real, with a co-founder who teaches computer science at City St George's, and the Innovate UK grants and the publicly described benchmark suggest a company willing to show its working.
Three reservations deserve equal weight. The performance figures are impressive but self-declared, measured by the vendor on a dataset it built itself, so they are a claim awaiting independent confirmation. The commercial opacity is total: no price appears anywhere, and the site publishes no email address at all, which means every question, including a request to delete uploaded files, travels through a contact form. And the data governance stops short of what a procurement team usually asks for, with no data processing agreement, no subprocessor list, no retention period and no stated hosting country, on a service whose whole purpose invites people to upload sensitive material such as legal files, CVs and evidence.
The company is refreshingly clear that no detector reaches complete accuracy and that its output belongs inside a broader verification process rather than standing as proof. That honesty is the right frame for using it. For an institution that already has a review procedure and simply needs a better signal across several media at once, the free entry point makes it easy to test on real content before any commitment.
- Choosing a selection results in a full page refresh.
- Opens in a new window.