Fraud used to scale badly. Every victim needed a person on the other end of the line: someone to write the email, hold the call, keep the story straight across weeks, and talk the target through the transfer. The cost of that labour was the natural ceiling on how far an operation could reach, and it was why the sophisticated schemes stayed rare and the mass schemes stayed crude.
Artificial superintelligence removed the ceiling. The scale and the sophistication shifted at the same time, and that is what moved fraud into the top tier of threats, next to ransomware and the nation-state intrusion.
What changed
Three things, together.
The operator is no longer human. Hyper-humanised agents now hold a video call, read the target's expression, match their tone, pause at the right moment and answer a hard question without breaking character. Tavus is a legitimate product and the clearest example of what the capability looks like: photorealistic conversational video agents, built from a couple of minutes of footage, that see, hear and respond in under a second. That technology exists for recruiters, tutors and support desks. The same class of capability, in the wrong hands, is a fraud operator that never tires, never slips, and runs a thousand conversations at once.
The proof is no longer real. A target who checks will find what they expect to find: a company website, a LinkedIn history, product reviews, a social media presence with years of posts, press mentions, a translated contract. Superintelligence generates all of it, consistently, in any language, faster than anyone can verify. The checks people were taught to do now confirm the lie.
The victim cannot tell. Audio and video fraud is now good enough that the average person, on a normal phone or laptop, cannot distinguish a superintelligent agent from a real human. The instinct that used to protect people, that something feels off, has nothing to work with.
The numbers
The FBI's Internet Crime Complaint Center recorded $20.9 billion in reported losses for 2025, a 26 percent increase on the year before. Business email compromise, the corporate impersonation scheme in which an attacker poses as an executive, a vendor or a counsel to redirect a payment, accounted for $3.04 billion of that on its own. For the first time the report carried a dedicated section on artificial intelligence, noting cloned executive voices authorising transfers and fabricated video used in BEC, romance, employment and investment schemes.
Those are reported losses from victims who recognised what happened and filed. The real figure is larger.
Two victim populations, one capability
For the enterprise, the threat is spoofing and phishing at superintelligence scale: a vendor whose domain is one character off, a CFO on a video call who is not the CFO, a new hire who passed every interview and is not a person.
For the individual, the threat is investment and romance fraud: recruitment into an investment club, advice from a mentor who does not exist, a trading platform whose reviews were written by the operation that runs it, a relationship built over months by an agent that remembers everything.
The techniques are old. Superintelligence elevated every one of them, by taking the human out of the operator seat and putting one into the victim's screen.
Guardian's answer: verify the human, govern the machine
Fraud at this scale is an identity problem, on both sides of the screen.
ScreenComply establishes that the human is real. On a hiring call, a client onboarding, a high-value approval or a remote session, ScreenComply verifies that the person on the other side is who they claim to be and is acting unassisted, not an agent, not a replica, not a person reading from one.
Monolith defends the layer an impersonation campaign depends on. Lookalike domains, registrar and DNS changes, certificate issuance, mail authentication, the identity and credentials a synthetic employee would use once hired, the egress path a planted agent would need: Monolith treats all of it as attack surface to inventory, validate and act on, and it does so continuously rather than after the loss. The hyper-humanised agent page covers what that looks like against voice and video fraud specifically.
Neither answer is a training poster. The era in which a careful person could spot the fraud is over; the control has to be structural.
How an in-line hardware layer turns all of this into an enforcement problem rather than a judgement problem: stopping social engineering at the hardware layer.
Frequently asked questions
What is meant by fraud at superintelligence scale?
Fraud operations in which artificial superintelligence performs the roles that previously required a human: holding conversations by text, voice or video, generating the supporting evidence a victim would check, translating in real time and running thousands of victims in parallel. The effect is that sophisticated, personalised fraud can now be delivered at mass-market volume.
How much did corporate victims lose to impersonation fraud in 2025?
The FBI IC3 report for 2025 attributes $3.04 billion in losses to business email compromise, the corporate impersonation category, out of $20.9 billion in total reported cybercrime losses. Both figures count only what victims reported.
Can a person reliably tell a superintelligent agent from a real human on a call?
Not any longer, on ordinary equipment. Conversational video agents now respond in under a second with natural expression and timing, and cloned voices pass casual listening. Detection has to be structural, through verification of the human and governance of the machine, rather than relying on the victim's instinct.
Is Tavus a fraud tool?
No. Tavus is a legitimate platform for conversational video agents used in recruiting, education and customer support. It is referenced here as the clearest public demonstration of how convincing a hyper-humanised agent has become, which is the same capability fraud operations now possess.
How do ScreenComply and Monolith work together against fraud?
ScreenComply verifies the human: that the person on a screen is real, is who they claim, and is acting unassisted. Monolith governs the machine side: the domains, identities, credentials and network paths an impersonation campaign needs to reach and act inside your organisation. Together they cover both sides of the screen.