Frequently asked questions

Diopter is used by organizations where a convincing impersonation can move money, grant access or change an identity record, which in practice means financial services, enterprises with exposed executives, and any team running remote identity checks.

Inside those organizations the users are usually the people at the point of contact rather than the security team alone. Help desk agents resetting credentials, finance teams approving payment changes, and recruiters running remote interviews all face the same problem, which is deciding whether the person on the call is real while the call is still live.

Security and fraud teams sit behind them, using the same signals to understand how an attack developed and where else it may have been attempted.

The matching playbooks are help desk defense, financial wire fraud and candidate fraud.

Synthetic speech carries artifacts that human speech does not, and Diopter listens for them while the call is in progress rather than analyzing a recording afterwards.

Cloned voices are generated by models that reconstruct speech rather than produce it physically, so they leave traces in micro-timing, spectral structure and the way breath and pacing sit against the words. Diopter scores these continuously through the call, which matters because an attacker may only reveal the cloned voice at the moment of the request.

Voice is also assessed alongside the behavior of the conversation itself. A cloned voice asking for an urgent payment change is a stronger signal than either observation on its own, and that combination is what separates detection from a simple audio classifier.

See how audio deepfake detection works or read the guide to deepfake audio detection methods.

Remote onboarding asks a stranger to prove who they are through a camera and a microphone, which is precisely the surface a real-time generation model is built to defeat.

Diopter ties identity to the action rather than checking a face once at the start of the session. It compares the person on the call against trusted identity signals, including the context and the channel the request arrived through, and it scores video and audio continuously for face-swap and voice-cloning indicators as the session runs rather than at a single capture moment.

The second half of KYC is the instruction that follows the verification. Diopter validates payment and account details for fraud signals such as a brand-new email domain or a freshly ported number, and it flags requests that route around your normal approval and verification controls. The output your team sees is a hold for verification with the reasons attached, so whoever is handling the case knows what to check next.

See payment and identity verification, and AI fraud detection in financial services for the wider onboarding picture.

A remote interview hands an attacker a controlled channel and a script, so Diopter scores that channel while the interview is running rather than after an offer has gone out.

Diopter scores the candidate’s video and audio for deepfake and voice-cloning indicators during the screen, and it checks the person applying against trusted identity signals before access and payroll are granted. It also looks across the whole loop rather than at a single call: the same backing actor or persona appearing across interview rounds is a far stronger signal than any one frame, and it is the pattern that exposes a proxy feeding answers to the candidate on screen.

Behavioral signals sit alongside that, including the evasion and coaching patterns common to fraudulent interviews. When those combine, Diopter flags the interview before an offer is extended, which is the last point at which your recruiting team can act cheaply. Gartner projects that one in four job candidates will be synthetic by 2028, so this is a screening problem rather than an edge case.

More detail on candidate fraud detection, and our write-up of deepfake job interviews.

An attacker impersonating your CEO or CFO is not trying to pass a media test. They are trying to force an approval, so Diopter scores the media and the play together.

Diopter scores the executive’s video and voice for deepfake and cloning indicators on the live call, then tracks the authority framing, manufactured urgency and isolation that define an impersonation attempt: the confidential deal, the closing window, the instruction to keep it between the two of you. It also flags approvals and access requests that fall outside your normal controls, which is what turns an uneasy feeling into something your finance or security team can act on.

The verdict is a hold for verification, raised while the call is still running rather than in a report afterwards. Timing is the whole point here, because losses in this category land inside a single call: $25.6 million on the deepfake CFO video call at Arup in 2024.

Attackers can clone your CEO’s face and voice. They cannot clone the absence of a real working relationship, and that is the layer Diopter reads. See executive impersonation defense and our breakdown of the Arup deepfake scam.

Book a walkthrough through the contact form and you will get a 30 minute session, run under NDA, with no requirement to install anything first.

The session works best when it follows your own threat surface rather than a generic script, so it helps to arrive with the scenario that concerns you most, whether that is a help desk credential reset, a payment change request or a remote interview.

If you are still comparing options, the free detectors on the site let you test video, audio and image detection immediately without speaking to anyone.

Diopter is built for enterprise deployment, and the questions that usually decide an enterprise purchase are about deployment and data rather than about the detection claim itself.

On deployment: on-prem and hybrid deployments are supported, capture works either with a meeting bot or bot-free, there is no caller-side install to push to the people you are protecting, and wider rollout runs through MDM including Intune and Jamf. On data: retention is configurable, including a zero data retention option, so sensitive call media can stay inside your perimeter. Diopter is SOC 2 Type II in progress rather than certified, and that is worth checking against your own procurement threshold before you shortlist.

Reliability in this category is mostly a question of how a system behaves under imperfect conditions. Calls arrive compressed, connections drop, and audio quality varies by region and device. Diopter reads video, audio and conversational signals together, so degradation in one of them does not remove the verdict.

The other half of reliability is operational. Detection has to reach the people handling the call while the call is still happening, which is why Diopter scores in real time rather than producing a report after the fact. Our own guide to evaluating deepfake detection tools sets out what to test before you commit, and enterprise deepfake prevention covers the rollout side.

Email support@diopter.ai and you reach the team that handles platform usage, investigations, integrations, account administration and security-related inquiries.

It is worth including the detail that lets someone act without a round trip: your organization, the environment or tenant involved, when the issue occurred, and whether a live call or an open investigation is currently blocked by it. If your request concerns a suspected attack in progress, say so in the subject line, because that is what separates an operational emergency from a configuration question.

If you are not a customer yet and the question is about evaluating Diopter rather than running it, the contact form is the faster route, since it reaches the team that runs walkthroughs rather than the support queue.

Support here is treated as part of the detection product rather than as a separate desk, because most requests arrive attached to a call that has already happened and a decision someone still has to make.

That shapes what the team is set up to handle. Platform usage, investigations into a specific flagged call, integrations, account administration and security-related inquiries all reach the same address rather than being routed through tiers you have to navigate first. Requests tied to a live or time-sensitive operational or security matter are prioritized ahead of configuration and administration work, because a verdict your team cannot act on is worth very little after the fact.

Diopter does not publish response-time commitments or named support tiers, so if a contractual SLA matters to your procurement process, raise it during the walkthrough rather than assuming a default.

Every request goes to one address, is reviewed by the team that owns the area it touches, and is then ordered by priority and impact rather than by the time it arrived.

In practice that means a suspected attack in progress or a blocked investigation moves ahead of an account administration change, and a question about an integration reaches the people who built it rather than a general first line. Platform usage, investigations, integrations, account administration and security-related inquiries all run through this same route.

What speeds a request up is giving the reviewer enough to locate the event: the call or session in question, when it occurred, the environment involved, and what your team saw against what they expected to see. Diopter does not publish a response-time commitment, so treat the urgency you signal in your subject line as the mechanism that surfaces a time-critical request.