Amazon Bedrock AgentCore expands to two regions and moves payments to GA

AWS expanded Amazon Bedrock AgentCore to two new regions — US West (N. California) and Asia Pacific (Hyderabad) — allowing customers to build and run AI agents nearer to their users for reduced latency. The platform bundles agent runtime, identity and access control, policy management, session persistence, tool connectivity, evaluations, and observability, with security enforced at the infrastructure level and support for any framework and any model.

More consequential is AgentCore payments reaching general availability. The feature lets agents autonomously discover, access, and pay for APIs and MCP tools, using x402 protocol support and Coinbase credential provisioning to settle transactions — a concrete step toward agents that transact on their own rather than merely retrieve information.

The releases fit AWS's aggressive agent push this week, which also included extending Bedrock Guardrails to tool interactions via the Strands Agents SDK, integrating Amazon Redshift with an Agent Toolkit for Claude Code, Kiro, and Cursor, and adding Grok 4.6 to Bedrock. AWS's own X account even asked whether 'AI agent sprawl is the new technical debt,' hinting at the governance tension its tooling aims to address.

The autonomous-payments capability raises obvious risk questions — spending controls, fraud, and auditability when agents move money — which is precisely why the guardrails and identity pieces ship alongside it. Watch enterprise adoption, how spend limits and human-in-the-loop approvals are configured, and whether x402/Coinbase settlement sees real production usage beyond demos.