How Forter Stops AI Agent Fraud

The same capabilities that make AI tools useful for shoppers make them powerful for large-scale fraud and abuse: speed, scale, and the ability to mimic human behavior. Our dark web intelligence shows fraudsters actively discussing and sharing ways to put these tools to work.

Over the past two months, Forter has seen automation-driven fraud increase 2x. This isn’t a blip. It’s an early signal of where fraud is heading as agentic commerce scales. And as legitimate agent traffic grows, bad actors will try to blend into it. When “a bot is buying this” stops being unusual, bot activity stops being an obvious red flag.

Blocking all automated traffic isn’t the answer. It would shut the door on the fastest-growing channel in commerce. Instead, merchants need to understand the distinct ways bad actors are harnessing agents to ensure agentic commerce trust.

Two attack vectors to watch

Forter is seeing two particular attack vectors shifting quickly in this environment. 

• Malicious agents. Bad actors build their own malicious AI agents, powered by stolen identities and payment credentials, sometimes running them on otherwise trusted platforms.
• Agent takeovers. The agent is legitimate, but the person controlling it is not. Fraudsters hijack a real buyer’s agent to capitalize on the reputation and trust that buyer has built over time.

How Forter is staying ahead and building agentic commerce trust

Creating agentic commerce trust and protecting merchants in this environment takes several layers working together.

See agents even when they don’t announce themselves. Despite several identification protocols being published, several agents choose to remain anonymous. We apply our cyber intelligence to detect agentic activity when it isn’t declared, rather than relying on a single platform’s declaration or allow-list. We also work with providers to enable data sharing for fraud detection, including through our involvement in emerging standards like AP2 and TACP.

Match detection to the threat. Every attack vector calls for its own detection approach, and we tailor our defenses accordingly. Take the two vectors above as examples. For fraudster-created agents, we assess plausibility signals such as velocity, preferences, delivery locations, and identifiers across our network. Furthermore, we validate the consistency of the consumer agent over time. For example, if a consumer usually uses Muse, and we start to see purchases on Instinct, we would evaluate that as a potentially concerning signal. For agent takeovers, we compare the live session against the owner’s historical baseline and flag sudden shifts in behavioral patterns such as order frequency, shipping destinations, and more.

Tune models to the channel. Agentic commerce changes the signals. Some device and connection data is lost when an agent transacts, but that’s not new. Phone orders have always worked this way. Reliable agent detection lets us run machine-learning models tuned to the characteristics of each channel, and even each agentic platform, instead of a one-size-fits-all model.

Here’s a simple example. On a standard web transaction, a purchase from a remote server might be a suspicious indicator. On certain agent platforms, running on a remote server is expected for every legitimate actor. A generic model would misread that signal and decline good customers. A tuned model won’t.

The network advantage

All of this depends on knowing what “normal” looks like for each person, and that’s where Forter’s network matters. Because we see behavior across merchants, we can verify that a purchase is consistent with a person’s past behavior and preferences: identity details, transactional habits, even how they use agents. That context doesn’t exist on any single merchant’s site.

The same visibility lets us spot new fraud vectors that exploit AI agents and shut them down quickly, often before they reach most merchants.

The bottom line

Agentic commerce will be enormous, but only if consumers and merchants trust it. Agentic commerce trust depends on fraud prevention that can tell good agents from bad ones, understands each channel’s quirks, and keeps a real human accountable behind the automation.

Fraudsters are already adapting. Forter can help merchants enable Agentic Commerce, while keeping the fraudsters out.