American Express unveils intent framework to govern AI agents

Amex’s framework is one of the first concrete attempts to redesign payment rails for a world where AI agents, not humans, click the “buy” button. By separating a cryptographically signed record of user intent from the payment credential that an agent actually uses, Amex is creating a control layer that can be audited, revoked and constrained as agents grow more autonomous. That is crucial if we are headed toward personal AIs that can reorder goods, negotiate subscriptions or book travel with minimal human oversight.

Strategically, this move positions traditional card networks as governance infrastructure for agentic commerce, rather than passive pipes behind Big Tech wallets. If Amex’s approach spreads to Visa and Mastercard or gets picked up by large merchants, it could become a de facto standard for how autonomous agents transact, defining where liability sits when things go wrong. For the broader race to AGI, it is another signal that the surrounding ecosystem of identity, risk and compliance is starting to adapt to agentic systems. That makes it easier for companies to roll out more capable, semi‑autonomous agents, knowing there are at least some guardrails between model hallucinations and real‑world financial damage.