A$11.4M (~US$8.15M) · Apate.AI + 2 more

What they're building

Apate builds conversational AI agents whose entire job is to be a convincing scam victim. They answer the scam call or the scam text, hold a realistic multilingual persona, and stay in the conversation — some run as long as two hours. The name is the point: Apate is the Greek deity of deceit, and the company's premise is that deception is a defensive weapon, not just an attacker's tool. The product is not a blocker, and that is the interesting part. Every conversation is an instrumented sensor. While the agent wastes a scammer's most expensive resource — human time on the phone — it is extracting the things that actually identify a fraud network: mule account numbers, crypto wallet addresses, the scripts being read, the social-engineering patterns being tested this week. Those get turned into structured intelligence artifacts and fed back into an institution's fraud-defense systems. Commonwealth Bank has run more than 2.5 million autonomous conversations on the platform and pulled over 250,000 intelligence artifacts out of them. The thesis underneath is that anti-fraud has been a passive, post-hoc discipline. You blocklist a number after it has already worked, you reimburse a customer after the money has already gone. Apate inverts the direction of travel: engagement itself becomes the collection mechanism, and the defender gets to run the experiment. Its site names Commonwealth Bank, TPG Telecom, Microsoft, Macquarie University and Australia's Office of National Intelligence among the organisations it works with, which tells you the buyer is a bank, a telco, or a government — the three places where scam calls actually terminate.

Why this matters

Startup Daily's coverage of the round puts the global scam industry at around $1.4 trillion a year. Whatever the precise figure, the structural shift is unambiguous: cheap synthetic voice and cheap LLM text have collapsed the marginal cost of running a scam, which is why volume has exploded. The under-discussed corollary is that the same collapse runs in the defender's direction. Before generative voice, a honeypot that could hold a scammer on the line for two hours meant paying a human to sit there for two hours, which is why nobody did it at scale. Now the economics reverse: the defender can field thousands of simultaneous convincing victims for nearly nothing, and every one of them is a listening post. That is a genuinely new capability, not an incremental one, and it is the reason a Sydney research spin-out is suddenly being funded to open in the US and Europe at the same time.