Apate Sends Autonomous Voice Bots Into the Line of Fire

What happened

Most founders pitch conversational AI to replace call center workers. Dali Kaafar built an army of software decoys designed to take a beating from telephone racketeers instead.

His Sydney startup, Apate, emerged on stage at Moscone West during TechCrunch Disrupt in San Francisco with a simple proposition. Instead of screening fraudulent calls, enterprise telecom carriers and retail banks route suspicious inbound traffic straight into automated honeypots. Apate runs thousands of voice and text bots simultaneously. They sound like confused retirees, distracted executives, or gullible retail investors. They speak across multiple languages, simulate varied regional accents, and mirror natural human speech delays.

The synthetic personas do not just waste time. They extract actionable counterintelligence. The bots stay on the line until the boiler room operative gives up banking coordinates, secondary burner numbers, crypto deposit addresses, and targeted brand identities.

The raw numbers show real operational scale:

• Apate has absorbed 2.5 million fraudulent interactions to date.
• The synthetic fleet has burned over 17 cumulative years of fraudster call time.
• Australian telecom provider TPG Telecom diverted nearly 590,000 suspect calls into Apate honeypots.
• Enterprise client Commonwealth Bank of Australia logged a 76 percent decline in publicly reported fraud losses.
• Another unnamed Australian banking client deployed the software to track down and shutter 22 active syndicates.

The venture market noticed. In August 2026, Apate banked an $8.15 million seed round led by institutional capital including Lobby Capital, OIF Ventures, Investible, Concept Ventures, and Baobab Ventures. That injection brought total balance sheet funding just under $10 million for the 20-person operation.

The background

The business started with an interrupted lunch. Kaafar, then an academic researcher at Macquarie University, sat at a family picnic when a fraudulent caller rang his mobile. He decided to play along to entertain his kids. He kept the operative spinning wheels for 44 minutes. The family got a laugh out of it, but Kaafar walked away with an operational calculation. That phone call occupied nearly three-quarters of an hour where the criminal could not dial an actual victim. It was effective, but manual scam-baiting does not scale.

Kaafar pulled the problem into Macquarie University labs. His research team applied natural language processing and machine learning pipelines to thousands of recorded telephone shakedowns, reverse-engineering how callers navigate social engineering scripts. In 2024, they spun the intellectual property out into Apate, naming the corporate entity after the ancient Greek deity of deception.

The commercial timing matched an explosion in synthetic crime. United States federal records show the scope of the problem. In 2025 alone, the FBI logged over 22,000 formal complaints regarding crimes powered by machine learning, generating roughly $893 million in documented victim losses. Rogue investment schemes accounted for more than $632 million of that haul. Criminals used voice cloning models to fake the cadence of company executives and prominent figures while automated bots ran social channels.

For two decades, scam-baiting was an eccentric internet hobby. YouTubers and forum posters spent afternoons baiting overseas call centers, tricking them into reading classic novels or wasting courier fees. That was guerrilla theatre. Apate turned it into an enterprise telemetry pipeline.

Banks and telecoms spent billions trying to solve the problem backwards. Carriers blacklisted phone numbers, but spoofing platforms generate millions of fresh VoIP caller IDs in seconds. Financial institutions built heuristic fraud engines to flag anomalous transfers, but an alert triggers only after the customer hits send. Apate intercepts the interaction before the wire leaves the building.

What people are saying

Apate co-founder and chief commercial officer Brad Joffe is unusually blunt about why the legacy defensive perimeter failed.

"Most of the industry works after the fact. Telcos block numbers, but scammers just get new ones. Banks watch transactions, but by then the scammer has already reached the customer. We go to the scammer directly and get the information before anyone gets hurt."

Joffe points to the intelligence payload as the real product. Apate does not just terminate calls. The platform combines commercial frontier foundation models with proprietary models fine-tuned on Kaafar's academic research and ongoing call logs. When enterprise security desks log into the Apate intelligence portal, they run directed search queries against active fraud campaigns.

A corporate security director can spin up targeted conversational agents with a single operational objective. If an ongoing phishing wave impersonates a specific retail brand, the agent's prompt instructs it to play along until the caller provides a payment drop address. If the target is an investment scam on an open messaging board, text bots scrape community groups, introduce themselves, adopt tailored background personas, and trigger private direct messages until the operator serves a malicious download link or a crypto wallet address.

Internal performance metrics at Apate do not look like standard enterprise SaaS dashboards. The engineering team monitors caller vitals, average call duration, and explicit language. When an angry boiler room worker starts screaming profanities into the line, the Apate office considers the prompt a success. It proves the model held the fraudster captive until their operational quota collapsed.

What happens next

The immediate corporate agenda is geographic expansion. Apate used its Startup Battlefield 200 appearance at TechCrunch Disrupt in San Francisco to introduce the platform to North American enterprise buyers. The newly secured capital will fund two primary footholds: opening a corporate sales office in London and establishing direct operations inside the United States.

The engineering footprint will have to scale quickly. Synthetic boiler rooms are not static targets. Criminal rings are actively deploying open-weight voice models, running their own low-latency speech agents to automate initial cold calls before patching in human closers. That changes the technical battlefield. When Apate runs its bots in 2027, it will increasingly pit defensive neural nets against offensive synthetic callers.

Success creates distinct operational bottlenecks for the startup:

• Compute overhead climbs rapidly when thousands of concurrent agents process sub-second voice synthesis.
• Fraud networks regularly alter their call trees to identify synthetic speech artefacts and latency pauses.
• Regulatory boundaries across Western markets restrict how telecommunications operators divert consumer voice traffic into private machine monitoring arrays.

For now, the economic model holds up. Every hour a fraud syndicate spends talking to an Apate server is an hour its payroll produces zero revenue.