Reality Defender

• Identity-verification platforms such as IDmission, which runs Reality Defender inside its onboarding pipeline, already receive the selfies it scores, so any of them could replace its signal with an in-house detector.
• New generative models can lower the accuracy of existing detectors, so Reality Defender has to keep sourcing fresh synthetic training data, as its ElevenLabs partnership does for voice.
• The academic benchmark in the reviewed sources that tested Reality Defender’s models anonymized each vendor’s results and found no commercial model at 90% accuracy, so buyers cannot compare its accuracy with rivals’ from that study.
• Orange Business, AU10TIX, and ZeroFox have agreed to embed the detector in their own services, and they own those customer relationships. A partner could replace Reality Defender for customers who receive its detection solely through that partner.
• The SAM.gov record of the DHS purchase lists only a special notice of intent to buy sole-source, so the federal reference would weaken if no award follows.

Reality Defender addresses fraud that uses AI-generated voices, faces, and images to pass checks built for real people. Its own anonymized case study describes a tier-one bank whose deepfaked calls got past voice biometrics and callback checks. Its IDmission case study describes an AI-generated selfie that cleared document and face-match checks during onboarding.

The problem is documented outside the vendor. In April 2024 a Senate Judiciary subcommittee heard testimony on election deepfakes, with CEO Ben Colman among the witnesses. An academic benchmark found that no commercial detector it tested reached 90% accuracy.

The scale estimates come from an independent market report. Biometric Update and Goode Intelligence forecast deepfake fraud attempts rising from about 145 million in 2026 to nearly 335 million in 2028. They expect voice detection checks to reach about 5.46 billion. Reality Defender’s named buyers are bank contact centers, identity-verification platforms, and government agencies. [s9, s10, s29, s27, s22, s1]

Reality Defender sells one detection platform with separate models for audio, video, and images, delivered through four products. RealCall flags cloned voices and AI callers during live calls, and RealMeeting alerts hosts inside Zoom and Microsoft Teams. RealAPI returns structured results to a customer’s own application, and RealScan is a browser app for uploading files.

Each product returns a detection result for the customer’s own systems to act on. RealScan labels every file Authentic, Suspicious, or Manipulated with supporting detail. The contact-center page tells buyers to let their own IVR and fraud rules decide what happens next. The company calls itself detection-only, returning a signal to the system that owns the decision.

Deployment options suit regulated buyers. RealScan’s enterprise tier lists on-premises, private cloud, containerized, and air-gapped laptop deployments. RealMeeting states that no personal data, recordings, or voiceprints are stored.

Reality Defender’s technical work is documented beyond its marketing pages. The company holds a granted US patent on audio spoof detection, filed in 2024, with another application listed as active. Its research index lists papers such as Alethia, an encoder for voice deepfakes, dated April 2026. [s2, s21, s6, s7, s8, s10, s13, s34, s15]

Reality Defender positions itself as an independent detection layer that feeds other security and identity systems. Its 2026 partnerships follow that shape: AU10TIX announced in July 2026 that it will evaluate the Reality Defender signal alongside its own identity-verification indicators, and ZeroFox added the detector to its external threat workflow.

Independent parties place Reality Defender among dedicated deepfake detection vendors. CB Insights lists Sensity among its competitors, and an academic benchmark tested its models alongside Hive, Pindrop, and Sensity AI. Reality Defender says Gartner named it a Market Shaper in a June 2026 Emerging Market Quadrant for deepfake detection startups.

Those partners are also potential rivals. Identity-verification platforms such as IDmission already receive the selfies Reality Defender scores, so one of them could build a detector of its own. [s19, s31, s28, s27, s15, s10]

Reality Defender’s customer evidence includes customers speaking for themselves. A J.P. Morgan CIO, quoted in Reality Defender’s release on the bank’s 2025 Hall of Innovation award, said the company helped secure its communications against deepfake impersonation. IDmission’s CEO is quoted in a Reality Defender case study about a deepfake selfie stopped during onboarding.

Other customer names come from AWS and from Reality Defender itself. An AWS Startups article says it serves Accenture, Booz Allen Hamilton, tier-one banks such as JP Morgan, and US and UK government agencies. Reality Defender’s anonymized case study reports a global bank running more than 1.7 million calls through the product in 2024.

The 2026 deals widen distribution through partners and government. Biometric Update reported the Orange Business agreement in April and the AU10TIX partnership in July. In September KPMG took a minority stake and said it would bring the detector into its cyber and fraud practices. The same month, DHS posted a notice of intent to buy the platform sole-source.

Reality Defender does not disclose revenue or a customer count, so its scale is visible through the named accounts and deals. Biometric Update cites a LinkedIn post in which the company claimed its strongest quarter to date. [s24, s10, s26, s9, s20, s19, s17, s30, s18]

CEO Ben Colman brings commercial security experience and public standing in the deepfake field. He led cybersecurity commercialization at Goldman Sachs and worked with Google’s partnerships team. A Senate Judiciary subcommittee called him as a witness at its 2024 hearing on election deepfakes.

The team’s technical work is on the patent record. Colman, co-founders Ali Shahriyari and Gaurav Bharaj, and two other researchers are named inventors on Reality Defender’s audio spoof-detection patent, and the company publishes a continuing research series. Reality Defender won the 2024 RSA Conference Innovation Sandbox.

The operating team extends beyond the founders. The company page lists a COO, CTO, and CRO alongside Colman, and Shahriyari and Bharaj appear as co-founders on its advisory council. Reality Defender says its detection platform was first developed inside a nonprofit research initiative, and the company was founded in 2021. [s4, s29, s13, s5, s34, s32, s3, s26]

Reality Defender runs a Vanta trust center that lists SOC 2, GDPR, UK Cyber Essentials, and HIPAA. Its documents include a SOC 2 Type 2 report dated October 2025, a 2025 Cyber Essentials certificate, and a 2026 penetration test. It also lists a SOC 2 bridge letter and a certificate of insurance.

Deployment choices address buyers who keep media in-house. RealScan’s enterprise tier offers on-premises, private cloud, containerized, and air-gapped deployments, and RealMeeting does not store recordings or voiceprints.

These credentials ease a buyer’s security review, but a funded rival could obtain the same ones, so they do not keep a competitor out of an evaluation. No federal authorization such as FedRAMP for the platform appears in the reviewed sources. [s16, s6, s7]

Add analyzed competitors to compare them side by side with Reality Defender.

Reality Defender sells to organizations whose decisions depend on a voice, face, or image being real. Its homepage mentions financial institutions and government agencies, and its solution pages cover contact centers, identity verification, meetings, and media and evidence review.

Financial services is the most evidenced segment. A J.P. Morgan executive is quoted on the company’s work. An anonymized tier-one bank runs it in its contact centers. The CTO of that bank’s private client division describes it as a way to protect high-net-worth clients from impersonation. BNY invested in 2025, and KPMG, which plans to offer the detector to clients, invested in 2026.

IDmission runs the detector inside its onboarding pipeline. AU10TIX agreed in July 2026 to add it to its platform. In September 2026, DHS Homeland Security Investigations posted a notice that it intends to solicit the platform from a single source.

Telecom is another route to enterprise accounts. Orange Business, the enterprise division of Orange, agreed to integrate the detector into its video conferencing, contact center, and telephony services.

Reality Defender’s detection runs on separate audio, video, and image models that the company says it blends, tests, and updates. Audio models assess live and recorded speech, video models examine faces, frames, and sound, and image models inspect faces and full frames. The platform page describes models for those three media types, and the Orange Business deal adds document analysis.

The company holds a named, retained training corpus. Reality Defender wrote in March 2025 that ElevenLabs voice synthesis data, more than 295 hours of it, made up about 20% of its training dataset at that time. Its partners page lists ElevenLabs, Respeecher, and PlayAI as AI-safety partners supplying resources for its detection models.

Reality Defender has patented part of its method. Google Patents records US12189712B1, an audio spoof-detection method filed by Reality Defender in 2024 and granted in January 2025. The company also publishes research, including red-teaming work on audio detectors.

Independent accuracy evidence is limited. The academic benchmark in the reviewed sources anonymized each vendor’s results and found that no commercial model reached 90% accuracy. Reality Defender quotes Gartner’s description of an ensemble-of-models data network effect, and the Gartner note itself is not among the reviewed sources.

Reality Defender sells to large enterprises and government while adding channels through partners. Its customers include banks, with a J.P. Morgan executive quoted on the work, and an AWS Startups article lists Accenture, Booz Allen Hamilton, and US and UK agencies as customers.

In 2026 Reality Defender signed channel deals that put its detector inside other companies’ products. Orange Business agreed to integrate it into its communication services, AU10TIX announced plans to add it to identity verification, and ZeroFox added it to external threat protection. KPMG plans to bring it into its cyber and fraud practices after taking a minority stake.

A developer path runs alongside the enterprise sale. RealAPI and SDKs let developers embed detection, and the homepage offers 50 free audio or image scans a month.

Reality Defender publishes one self-serve price and keeps enterprise pricing custom. RealScan’s Business plan costs $399 with annual billing and covers 1,000 scans a month for one seat. The Enterprise plan lists custom pricing, custom scan volume, unlimited seats, and on-premises or private-cloud deployment.

A free developer tier lowers the cost of a first test. The homepage offers 50 audio or image scans a month at no charge through the API and SDKs.

Larger contracts start with a sales conversation. The contact-center and RealMeeting pages list no prices and invite buyers to request a demo.

Reality Defender delivers a score into the customer’s existing systems, and the customer makes the decision. The contact-center page tells buyers to let their IVR and fraud rules decide whether to verify, route, or end a call. In the IDmission case, IDmission owned the onboarding decision.

Real-time operation is documented in production. In the anonymized bank case study, the detector sends live alerts to about 600 frontline representatives. The bank first analyzed call recordings from Zoom, Meet, NICE, Twilio, and Five9. The case study reports more than 1.7 million calls analyzed in 2024, which are the company’s own figures.

Deployment adapts to regulated environments. Enterprise buyers can run RealScan on premises, in a private cloud, in containers, or on an air-gapped laptop, and RealMeeting keeps no recordings or voiceprints.

Reality Defender’s trust center, hosted on Vanta, lists SOC 2, GDPR, UK Cyber Essentials, and HIPAA. Its assessments include a SOC 2 Type 2 report dated October 2025, a 2025 Cyber Essentials certificate, and a 2026 penetration test. The trust center also lists a SOC 2 bridge letter and a certificate of insurance.

Privacy commitments appear on the product pages. RealMeeting states that no personal data, recordings, or voiceprints are stored, and the enterprise tier supports on-premises and air-gapped deployment.

A funded rival could earn the same credentials, so they ease a security review without keeping a competitor out. No federal authorization such as FedRAMP for the platform appears in the reviewed sources.

Reality Defender builds its reach through other companies’ platforms. ZeroFox added Reality Defender’s verdict to its external threat workflow, AU10TIX announced plans to add it to identity verification, and Orange Business is integrating it into telecom communication services.

Its partner program spans deployment, marketplace, professional-services, AI-safety, and technology partners. Accenture, Booz Allen Hamilton, Deloitte, IBM, and KPMG appear as professional-services partners, and AWS, Microsoft, and NVIDIA as technology partners.

ElevenLabs, Respeecher, and PlayAI are listed as providing resources for Reality Defender’s detection models, and the company said in March 2025 that ElevenLabs data made up about 20% of its training dataset.

Reality Defender’s leadership pairs a commercial security CEO with an in-house research team. Ben Colman led cybersecurity commercialization at Goldman Sachs before co-founding the company. The company page lists a COO, CTO, and CRO, plus directors of applied AI for vision and audio.

The founders’ technical work is on the patent record. Colman, Ali Shahriyari, Gaurav Bharaj, and staff researchers are named inventors on the granted audio spoof-detection patent. Shahriyari, a co-founder and advisor, previously served as director of product at the AI Foundation.

External recognition is specific to this field. The Senate Judiciary subcommittee called Colman as a witness on election deepfakes in 2024, and the company won the 2024 RSA Conference Innovation Sandbox. Colman told AWS that the company started as a nonprofit research project ahead of the 2020 election.