Frequently asked questions

Diopter AI is a deepfake and AI social engineering detection platform. It analyzes audio and video streams in real time to identify synthetic media, including voice clones, face swaps, and AI-generated identities, before they can be used to execute fraud. Diopter evaluates each call or session as a structured conversation event, not just a single audio or video artifact. Learn more about the platform

Diopter detects synthetic voice calls, deepfake video, AI-generated identities, and social engineering attack patterns. The platform evaluates audio and video streams for manipulation signals such as vocal pattern anomalies, facial coherence gaps, and behavioral indicators, and issues a verdict at each stage of the interaction.

Diopter is built for organizations where identity verification and high-trust communication are critical. This includes banks, fintech companies, KYC providers, media organizations, and enterprise security teams. CISOs, security operations leaders, and fraud prevention teams are the primary buyers.

Diopter serves financial services (banking and fintech), KYC and identity verification providers, media organizations, enterprise hiring teams, and organizations with executive protection requirements. Each vertical faces a distinct deepfake attack surface, from wire transfer fraud to synthetic candidate fraud in hiring. See industry playbooks

Diopter is built for security, fraud, IT, and operations teams across industries including financial services, real estate, insurance, mortgage, staffing, and recruiting.

Diopter is designed for organizations that need to protect high-stakes calls involving payments, approvals, hiring, and account access. Customer references and deployment use cases are available upon request through our Contact Us page.

A deepfake is AI-generated synthetic media, including video, audio, or image, that manipulates or replaces a real person’s likeness with sufficient fidelity to deceive. Deepfakes are produced using generative models trained on real recordings of the target. The term originally referred to face-swap videos but now covers any synthetic identity output: voice clones, AI-generated avatars, and composite personas used in fraud. Read our full guide to deepfakes →

Deepfakes are produced using generative AI models, typically GANs (Generative Adversarial Networks) or diffusion models, trained on video, audio, or image samples of the target. For voice cloning, as little as a few seconds of audio can be sufficient. For video deepfakes, face-swap models require more source material, but commercial tools have significantly lowered the technical barrier. Learn how deepfakes are built and detected →

Synthetic media is any audio, video, image, or text generated by an AI model rather than captured from the real world. It includes deepfake videos, AI voice clones, AI-generated avatars, and composite personas. In a fraud context, synthetic media is used to impersonate real individuals with sufficient fidelity to bypass human and automated verification.

Voice cloning is the AI-powered replication of a specific person’s voice from audio samples. A clone reproduces the target’s vocal tone, speech rhythm, accent, and characteristic phrases. Modern voice cloning models produce convincing outputs from as little as a few seconds of audio. Voice clones are deployed in fraud via phone calls, voice messages, and real-time call manipulation, impersonating executives, bank representatives, or known contacts.

A synthetic voice call uses an AI-generated voice, either a clone of a real person or a fabricated persona, to conduct a phone interaction. In fraud scenarios, synthetic voice calls impersonate executives, bank representatives, or government officials to authorize transactions, extract credentials, or manipulate employee behavior. The caller sounds indistinguishable from the impersonated individual to most listeners. Diopter detects synthetic voice in real time

Yes. A small amount of publicly accessible audio, such as a podcast appearance, a recorded video, or a voicemail greeting, is sufficient for modern voice cloning models to produce a usable replica. Executives, public figures, and anyone with a significant audio footprint online are particularly exposed. Voice cloning no longer requires specialized expertise or large audio datasets.

Diopter continuously analyzes live audio for synthetic voice indicators alongside identity signals and conversation context, helping identify cloned voices before critical actions are approved.

AI social engineering is the use of artificial intelligence, including voice cloning, deepfake video, and synthetic personas, to enhance traditional manipulation tactics. AI enables attackers to impersonate specific, known individuals with fidelity that bypasses human skepticism. The attack surface shifts from mass-targeted phishing to highly personalized, real-time impersonation of trusted contacts. Read our full guide to social engineering →

Vishing (voice phishing) is a social engineering attack conducted over voice calls. The attacker impersonates a trusted authority, such as an executive, bank representative, or government official, to extract credentials, authorize transactions, or manipulate employee behavior. AI voice cloning has made vishing attacks significantly harder to detect. The caller does not need to sound convincing; they sound exactly like the impersonated individual. How attackers use social engineering in voice fraud →

Authority framing is a manipulation tactic where the attacker asserts a position of power or legitimacy before making a request. In voice fraud, this means impersonating an executive, regulator, or law enforcement figure. The framing reduces the target’s likelihood of questioning or verifying the request through normal channels. Combined with voice cloning, authority framing becomes significantly more effective because the voice matches the claimed identity.

Deepfake detection analyzes audio and video streams for artifacts introduced by the synthesis process. For video, detection evaluates facial coherence, blinking patterns, lighting consistency, and boundary anomalies. For audio, it examines spectral features, micro-timing, and synthesis model fingerprints. Human detection is inconsistent; automated systems like Diopter evaluate structured signals across every frame and phoneme. Deep dive: how deepfake detection works →

Liveness detection verifies that a face or voice presented in a session belongs to a live human rather than a recording, photograph, or synthetic media output. It is a core component of KYC processes. Deepfakes have broken many first-generation liveness systems. Modern liveness detection must account for real-time synthetic generation, not only pre-recorded media attacks.

Yes. Real-time deepfake detection analyzes incoming audio or video streams continuously and issues verdicts with latency low enough to be actionable during an active call. Diopter is built specifically for this, running detection end-to-end across live calls on platforms like Zoom, Teams, and Google Meet. The challenge is maintaining accuracy under the compression and noise conditions of real-world calls, which differ significantly from controlled laboratory environments.

Deepfake banking fraud uses synthetic voice or video to impersonate account holders, executives, or relationship managers to authorize fraudulent transactions. Common patterns include cloned executive voices instructing wire transfers and synthetic identities bypassing remote onboarding checks. Banks face exposure at every voice-enabled touchpoint: call centers, remote verification sessions, and internal authorization calls.

KYC fraud using deepfakes involves presenting synthetic media, such as a real-time generated face, a video replay, or an AI-generated identity, to pass remote identity verification. As financial institutions move onboarding online, the attack surface for synthetic identity fraud has expanded significantly. First-generation liveness detection systems are increasingly insufficient against real-time generation models.

Deepfake hiring fraud involves synthetic candidates using cloned voices or video avatars in remote interviews to fraudulently obtain employment, access credentials, or infiltrate organizations. The attacker passes recruitment screening as a fabricated or stolen identity. Sectors with significant remote hiring and privileged system access, including technology, finance, and defense, are primary targets.

Diopter helps detect synthetic candidates and proxy interviewers by analyzing live video, audio, identity signals, and conversation patterns throughout the interview process.

Diopter helps strengthen remote identity verification by detecting synthetic audio or video and identifying manipulation patterns that may indicate attempted identity fraud.

CEO fraud is a social engineering attack where the attacker impersonates a company’s CEO or senior executive to instruct employees to execute unauthorized wire transfers, share sensitive data, or bypass standard verification procedures. AI voice cloning has made CEO fraud significantly more effective. Recipients receive a call that sounds exactly like their CEO, eliminating the skepticism that text-based impersonation would typically trigger. Understand the social engineering tactics behind CEO fraud →

Executive impersonation is the fraudulent representation of a C-suite leader, such as a CEO, CFO, or board member, to manipulate employees, partners, or financial institutions. It spans email-based BEC, synthetic voice calls, and deepfake video. The common thread is exploiting the authority of an executive identity to override normal approval and verification processes.

C-suite deepfake fraud typically follows a structured arc. The attacker researches the target executive’s voice from public recordings, clones it, then calls a financial controller or operations lead with an urgent wire transfer or data release instruction. The call uses the executive’s speech patterns, includes urgency signals that discourage verification, and completes before any fraud flag triggers. Documented losses from single incidents have reached into the millions. Diopter monitors the conversation arc as it builds

Diopter verifies identity, detects synthetic media, and monitors for authority, urgency, and other manipulation tactics that commonly appear in executive impersonation attacks.

Documented deepfake fraud losses include individual incidents in the tens of millions of dollars. One widely reported case involved a finance employee who transferred over $25 million following a deepfake video call impersonating company leadership. Aggregate losses from AI-enabled social engineering and synthetic identity fraud run into the billions annually. Actual figures are likely higher due to underreporting and misclassification as standard fraud.

Regulatory frameworks addressing deepfakes include the EU AI Act, which requires disclosure of synthetic media, and US state-level laws in California, Texas, and Virginia covering specific use cases. Financial regulators including the FFIEC and FCA have issued guidance on synthetic identity risk in digital onboarding. Compliance teams should monitor both technology-specific legislation and existing fraud statutes being applied to deepfake cases.

A deepfake risk assessment maps the organization’s communication touchpoints, including call centers, executive channels, remote onboarding flows, and financial authorization processes, against known deepfake attack vectors. It identifies where voice or video authentication is relied upon, where human judgment is the primary control, and where synthetic media would have the highest impact if undetected. The output is a prioritized exposure map, not a compliance checklist. Walk an attack arc with Diopter

Diopter is built for enterprise environments with on-device detection where applicable, configurable data retention, tenant-level controls, MDM deployment, and SOC 2 Type II in progress.

Customers can contact the Diopter support team at support@diopter.ai for assistance with platform usage, investigations, integrations, account administration, and security-related inquiries.

Support requests are reviewed by the appropriate team and addressed based on priority and impact.

Diopter maintains a support process designed to address customer inquiries efficiently while prioritizing time-sensitive operational and security matters.

You can book a 30-minute walkthrough to see how Diopter detects deepfakes, models conversation risk, and helps teams respond before high-risk decisions are made.