Current landscape

We are in the early stages of a structural shift in how digital payments work. For the past 30 years, every payment model has assumed a human initiating the transaction. Card networks, fraud models, compliance frameworks, and dispute resolution processes all presuppose a human at the other end of the wire. That assumption is beginning to fail.

AI agents — software entities that autonomously browse, select, negotiate, and pay — are entering commercial deployments at scale. Enterprise procurement agents that manage vendor payments. Personal finance agents that optimise subscription spend. Travel agents that book and reschedule in real time. Each of these represents a payment initiator that is not a human.

AI agents in payments

The payment behaviour of AI agents differs from humans in ways that are deeply problematic for existing fraud models. Agents are fast — they complete checkout flows in seconds. Agents are consistent — their navigation sequences are deterministic and therefore look anomalous compared to human variance. Agents are rational — they never make the small "mistakes" (hovering, backtracking, re-reading) that humans make and that fraud models rely on as legitimacy signals.

New risk vectors

Agentic commerce introduces risk vectors that did not exist in human-only payment systems. A compromised agent can drain accounts faster than any human attacker. An agent operating under a manipulated prompt can be redirected to make purchases the account owner never authorised. Agent-to-agent payment loops can generate synthetic transaction volume that distorts fraud models trained on human distributions.

Detection strategies

Effective detection requires classification before evaluation. An AI agent must be identified as such before risk scoring begins — because the signal set appropriate for scoring an agent is categorically different from the signal set for scoring a human. Agent classification relies on network-layer signals (ASN, datacenter IP ranges), HTTP behavioural signals (header ordering, timing distributions), and session entropy analysis.

Predictions

By 2028, we project that AI agents will initiate 18% of all digital transactions by volume. By 2030, the majority of B2B payments will be agent-initiated. This is not a threat to be defended against — it is an architectural reality to be designed for. The payment infrastructure that thrives will be the one that can serve human and AI principals with equal precision.