Every year I am asked what is coming next in fraud, and most years the honest answer is "the same attacks, a little sharper." 2027 is different. A handful of genuine shifts are converging at once, and together they change not just the tactics but the assumptions your defences are built on. Here are the five I am telling every operator to plan for, and the one thing that ties them together.
1. Agentic commerce goes mainstream
AI agents that shop and buy for real people are moving from curiosity to real traffic fast, with some forecasts putting the share of online transactions involving an agent near 40% by 2027. That breaks the old rule that automation equals fraud: a legitimate agent is a script acting for a paying customer. You will need to welcome good agents and refuse fakes — a decision behavioural signals cannot make for you. (See AI shopping agents at the checkout.)
2. AI-agent impersonation becomes a standard attack
As agents earn trust, bots will wear their names. Much of the traffic already claiming to be a named AI agent does not come from that operator at all. Allow-listing by user-agent becomes an open door, and verifying a declared agent against the network it truly comes from becomes a basic control. (See verifying AI agents by their network.)
3. Residential proxies become the default disguise
"Datacenter equals bot, home equals human" is over. Residential and mobile proxy networks now carry the majority of fraud traffic by some measures, specifically to defeat blocklists. Static lists of bad IPs age badly; live classification of the network, at the moment of the request, is what holds up. (See residential proxies are now the default.)
4. Synthetic identities and deepfakes defeat verification
AI-built synthetic identities are the fastest-growing fraud type, and deepfake selfies and injected video are passing the biometric checks meant to stop them. When the identity and the face can be manufactured, identity checks need an independent backstop the fraudster has to defeat separately. (See synthetic identity fraud and deepfakes beating KYC.)
5. First-party fraud keeps rising
The fraudster is increasingly your own customer. First-party disputes have roughly doubled their share of fraud cases in two years, and the only defence is evidence captured at order time. (See first-party fraud is now a third of all fraud.)
The thread through all five
Notice what these shifts have in common. Identities can be synthesised, faces can be deepfaked, behaviour can be mimicked, user-agents can be forged, and lists go stale — but every one of these attacks still has to connect from somewhere, and that network is a physical fact that is expensive to fake and cheap to read. The durable move for 2027 is to build on live network classification, combined with cheap corroborating checks like the card's issuing country. Fraudex is built on exactly that: a live verdict on any IP's network and type, with proxy, VPN and bot detection, plus BIN intelligence, in a single call. Start with the proxy detection API, the VPN detection API and the BIN lookup API.