Kairos provides facial recognition APIs and made the rare decision to withdraw from law enforcement contracts citing ethical concerns about mass surveillance and documented racial bias in facial recognition systems. CEO publicly advocated against police use of facial recognition technology. Focuses on commercial identity verification applications only. Emphasizes diverse training data and active bias mitigation. One of very few facial recognition companies to take a public ethical stance.
Interested in promoting your brand? Create yours free!
Launch Your Brand →Sign in to leave a comment
Sign In →Ran packet analysis on their SDK. Sends facial templates to servers in three different jurisdictions. Cross-border biometric transfers with zero user awareness.
What accuracy rates does this system actually achieve across different demographics? Have bias audits been conducted?
Facial recognition deployed without consent of the people being scanned. No opt-in, no notification, no recourse.
Real-time community metrics & activity
Total Views
0
Total Upvotes
0
Active Brands
0
Comments Today
90
No posts yet. Be the first!
Brands getting discovered...
nowCommunity growing daily
1m agoUnlock full product intelligence
Deep analytics, privacy reports, competitor comparisons, and expert reviews. Everything you need to make informed decisions.
Try SeekerPro Free for 14 Days→$15.99/mo after trial. Cancel anytime. No card charged today.
Run a free AI-powered privacy audit. Compare data practices, transparency, and user rights.
Want AI-powered analysis?
Compare this brand against competitors, get growth predictions, and export reports.
Get deep-dive reports on every company that touches your data. SeekerPro members see breach timelines, DSAR success rates, and risk scores before anyone else.
Get Started FreeAdd your brand to the Noizz catalog of 28,697 brands · updated daily
NexusBro
AI Website QA Audits
Is your website as strong as your brand?
Run a free QA audit. 125+ checks. Copy-ready fix prompt in 60 seconds.
125+
Checks
100K
Pages
0-100
Score
Used by supabase.com, vercel.com, github.com, and 30+ sites
Free to get started. No credit card required.
Explore NoizzID.me
ID.me provides identity verification for US federal and state agencies including the IRS, VA, and Social Security Administration. Maintains one of the largest private facial recognition databases in America with over 100 million verified users. Faced congressional scrutiny after CEO admitted using 1-to-many facial recognition, comparing each face against the entire database rather than just 1-to-1 matching. Users must submit government ID plus live selfie for biometric matching with excessive data retention.
Clearview AI
Clearview AI built the world largest facial recognition database by scraping 30+ billion photos from social media, news sites, and public sources without any consent. Sells access to law enforcement agencies worldwide for real-time facial identification. Fined and banned by privacy authorities in France, Italy, UK, Australia, Greece, and Canada. CEO claims first amendment right to scrape faces from the internet. If you exist online, Clearview probably has your face.
PimEyes
PimEyes operates a public-facing facial recognition search engine that anyone can use to find where a face appears online. Upload any photo and find every public occurrence across the internet. Extensively used by stalkers, abusers, and doxers despite company disclaimers. Polish-founded but now Seychelles-registered to evade privacy regulation. Charges $30-300 per month for unlimited face searches with no consent from people in the database.
Scandal alerts, breach notifications, DSAR deadlines, and protection guides. Join 2,400+ privacy-conscious professionals.
No spam. Weekly only. Unsubscribe anytime.
Tools trusted by thousands of privacy-conscious users worldwide
No card charged today. Cancel anytime.
Want unlimited access? Explore SeekerPro
SeekerPro gives you the full picture on every brand.
Facial recognition operates through a pipeline of several distinct stages. First, face detection algorithms locate faces within an image or video frame, typically using convolutional neural networks trained on millions of face and non-face examples. Once detected, the face undergoes alignment, where landmarks like eye corners, nose tip, and mouth edges are identified and the image is geometrically transformed to a standard orientation. The aligned face is then passed through a deep neural network, often based on architectures like FaceNet, ArcFace, or DeepFace, which produces a compact numerical representation called an embedding or faceprint, typically a vector of 128 to 512 numbers. This embedding captures the essential identity features of the face in a way that is invariant to lighting, expression, and minor pose changes. For identification, this embedding is compared against a database of stored embeddings using distance metrics like cosine similarity or Euclidean distance. A match is declared when the distance falls below a configured threshold. The entire process can run in milliseconds on modern hardware.
The accuracy of facial recognition depends heavily on conditions and demographics. In ideal conditions with frontal, well-lit, high-resolution images, top algorithms achieve false match rates below 0.001 percent. However, real-world accuracy drops significantly. NIST 2019 Face Recognition Vendor Test evaluated 189 algorithms from 99 developers and found that the majority showed higher false positive rates for African American and East Asian faces compared to Caucasian faces, with differentials ranging from 10 to 100 times worse depending on the algorithm. For one-to-many searches against large databases, false positive rates for Black women were as high as 10 times that of white men. These disparities arise from training data imbalances, algorithmic design choices, and the physics of image capture, as many camera sensors perform less well with darker skin tones. The real-world impact includes wrongful arrests, denied services, and discriminatory policing. While some algorithms have reduced demographic differentials, the problem persists across the industry and is compounded by the reluctance of many vendors to submit to independent bias audits.
Facial recognition templates, the numerical vectors derived from face images, are stored in various ways depending on the provider and use case. Some companies store templates in centralized databases alongside associated identity information, creating high-value targets for attackers. Others use encrypted databases with access controls and audit logs. Apple Face ID takes a notably different approach: the facial template is stored exclusively in the Secure Enclave of your device, a dedicated hardware security module, and is never sent to Apple servers or backed up to iCloud. The mathematical template itself is not reversible to a photograph, but research has shown that with sufficient computational resources, approximate face reconstructions from templates are possible. Template protection schemes like cancelable biometrics apply one-way transformations so the stored template is not the raw embedding, allowing reissuance if compromised. Homomorphic encryption enables matching against encrypted templates without decryption. Despite these techniques, many commercial deployments still rely on basic database encryption, and breaches of biometric databases, like the 2019 Suprema BioStar 2 breach exposing over a million fingerprints and facial records, demonstrate the ongoing risk.
Legal restrictions on facial recognition are expanding globally but remain fragmented. In the EU, the AI Act classifies real-time biometric identification in public spaces as prohibited with limited law enforcement exceptions, and GDPR treats facial templates as special category biometric data requiring explicit consent. Belgium and Luxembourg have enacted strict bans on biometric surveillance. In the US, Illinois BIPA requires informed written consent before collecting biometric identifiers and has generated over 5 billion dollars in settlements. San Francisco, Boston, Portland, Baltimore, and several other cities have banned government use of facial recognition. Vermont prohibited its use by law enforcement statewide. Washington state requires consent notices for commercial use. In other regions, King County in Washington state and the state of Virginia have partial restrictions. Brazil General Data Protection Law classifies biometric data as sensitive. Japan, South Korea, and India have guidelines but limited enforcement. China has no restrictions on government use but introduced some commercial consent requirements. The global trend is toward increased regulation, but most jurisdictions still have no specific facial recognition law.
Consent requirements for facial recognition vary by jurisdiction and use case. Under GDPR, processing biometric data for identification requires explicit consent under Article 9, meaning freely given, specific, informed, and unambiguous affirmative action. A general terms-of-service checkbox is not sufficient. The consent must specifically address biometric processing, state the purpose, identify the data controller, and inform the subject of withdrawal rights. Under Illinois BIPA, written informed consent is required before collecting any biometric identifier, including face geometry. The consent must disclose the purpose, retention period, and destruction schedule. Texas and Washington have similar but less stringent requirements. In most other US states and many countries, no specific consent is required for facial recognition. Law enforcement use typically operates without individual consent, relying on statutory authority or public interest justifications. For commercial use in unregulated jurisdictions, consent is often buried in lengthy privacy policies or considered implied by entering a premises with posted signage. The lack of consistent consent standards means that a facial recognition deployment that is illegal in one city may be completely unregulated across the border.
One-to-one (1:1) matching, also called verification, compares a face against a single stored reference to confirm identity. This is what happens when you unlock your phone with Face ID or pass through an e-passport gate: the system checks whether you are who you claim to be. The false match rate is relatively low because the comparison space is limited to one candidate. One-to-many (1:N) matching, also called identification, searches a captured face against an entire database to determine who someone is. This is used in law enforcement to identify suspects from CCTV footage by searching against mugshot databases or watchlists. The technical and ethical differences are substantial. With 1:N matching, the false positive rate scales with the database size: searching against a million faces produces far more false matches than searching against ten. This is why wrongful arrests based on facial recognition almost always involve 1:N searches. The computational resources required also differ dramatically. Privacy implications are fundamentally different as well: 1:1 verification confirms a claimed identity with user participation, while 1:N identification operates without the subject knowledge and can identify anyone in a database.
Retail facial recognition has expanded rapidly, often without customer awareness. Major use cases include loss prevention, where stores scan faces to match against databases of known shoplifters, and VIP identification, where luxury retailers identify high-value customers as they enter. Demographic analytics platforms estimate age, gender, and emotional state of shoppers to optimize product placement and advertising. Digital signage systems adjust displayed advertisements based on who is looking at the screen. Some retailers track customer movement patterns through stores using facial recognition to analyze foot traffic. Walmart, Walgreens, and Kroger have tested or deployed facial recognition systems. Rite Aid deployed the technology in hundreds of stores for loss prevention before being banned by the FTC in 2023 for inaccuracy and discriminatory deployment that disproportionately targeted women and people of color. In China, face-pay systems allow purchases by scanning your face at checkout. The lack of transparency is a central concern: most shoppers have no idea their face is being scanned, and few stores provide meaningful notice or opt-out mechanisms.
Protecting yourself from facial recognition involves a combination of physical, digital, and behavioral strategies. Physically, wearing sunglasses that block infrared light, wide-brimmed hats, and face coverings significantly reduces recognition accuracy. CV Dazzle, a project by artist Adam Harvey, demonstrated that specific makeup patterns can confuse detection algorithms, though modern AI systems are increasingly resilient to these techniques. Adversarial accessories like specially patterned glasses developed by researchers at Carnegie Mellon can cause misidentification. Digitally, minimize your facial data footprint by removing photos from social media, using avatars instead of real photos, and opting out of facial recognition databases like Clearview AI and PimEyes. Request deletion from any service that collected your biometric data. Avoid services that require face scans when alternatives exist. Use browsers and search engines that do not track images. Behaviorally, be aware of camera locations, understand that many surveillance cameras are not visible, and recognize that smart doorbells and other consumer devices feed surveillance networks. Ultimately, systemic protection requires legislation, as individual countermeasures cannot fully defeat a determined surveillance infrastructure.
Facial recognition and deepfake detection are related but distinct technologies with an ongoing arms race between them. Deepfake detection systems analyze facial images and videos for artifacts of synthetic generation, including inconsistent skin textures, unnatural blinking patterns, irregular head movement, lighting inconsistencies, and compression artifacts unique to generative models. Some detectors examine biological signals like pulse-related color changes in facial skin that deepfakes often fail to replicate. Microsoft Video Authenticator and tools from Sensity AI and Reality Defender provide detection capabilities. Current detection accuracy varies: on known deepfake datasets, top systems achieve over 95 percent accuracy, but performance drops significantly on novel generation methods and high-quality deepfakes. Facial recognition systems themselves are vulnerable to deepfake attacks where a synthetic face is presented to impersonate someone. Liveness detection countermeasures like requiring head movement, blinking, or reading random numbers help but can be defeated by real-time face-swapping tools. The fundamental challenge is that generative AI improves faster than detection methods, creating a persistent cat-and-mouse dynamic. Organizations increasingly combine multiple detection approaches rather than relying on any single method.
Face-based authentication is evolving along several trajectories. On-device processing is becoming the dominant model for consumer authentication, with Apple Face ID, Android biometric unlock, and Windows Hello all processing facial data locally without sending it to servers. This model provides convenience while mitigating the privacy risks of centralized biometric databases. Multi-modal biometrics combining face recognition with voice, gait, or behavioral patterns will increase accuracy and reduce the impact of any single biometric being spoofed. Privacy-enhancing technologies like homomorphic encryption will enable face matching against encrypted databases, and federated learning will train better models without centralizing facial data. However, the identification use case, determining who someone is in public spaces, faces growing regulation and public resistance. The EU AI Act, expanding US state laws, and civil society pressure are constraining mass facial identification. Emerging alternatives include palm vein scanning, iris recognition at a distance, and continuous behavioral authentication that identifies users by interaction patterns rather than static biometrics. The likely future is bifurcated: widespread consensual face authentication on personal devices alongside increasingly restricted use of face identification in public spaces.
Get deep-dive reports on every company that touches your data. SeekerPro members see breach timelines, DSAR success rate...
Learn MoreRun a full privacy and compliance audit on any website in 60 seconds. NexusBro scans cookie consent, tracker behavior, a...
Learn MoreStop wasting hours on manual DSAR filings and cookie consent management. BliniBot handles the busywork so your team can ...
Learn MoreAmazon Rekognition
Amazon Rekognition provides cloud-based facial recognition through AWS with documented racial and gender bias confirmed by independent researchers. ACLU studies demonstrated 28 false matches with members of US Congress, disproportionately misidentifying people of color. Amazon imposed a moratorium on police use in 2020 but the API remains available to any customer with an AWS account for any purpose.