Stripe deploys AI agents for compliance, cuts review time 26%

Key takeaway

• Stripe deployed an AI agent system on AWS to handle financial compliance reviews for its massive payment volume, reducing manual review time by 26 percent.

• The system breaks complex investigations into bite-sized tasks, with AI agents gathering research and human reviewers making final decisions, achieving over 96 percent helpfulness while maintaining full auditability and regulatory control.

3 Key Points

FAQ

How does Stripe keep humans in control of compliance decisions?

Agent responses are provided as supplementary information to human reviewers, who must ultimately answer each sub-task of the review. The system uses multi-layered decision checkpoints and configurable approval workflows, keeping humans in the driver's seat rather than relying outright on agent responses.

What technical approach does Stripe use for the AI agents?

Stripe built the compliance agent using the ReAct (reasoning and acting) agent framework on Amazon Bedrock. The agent dynamically gathers relevant signals through tool calls, with a closed-loop mechanism that grounds reasoning in actual data by forcing each tool output to be processed as an observation before the agent continues.