AWS Sharpens Its AI Tools: Cost Controls, New Claude Model, and Agent Management

For teams moving AI projects from prototype to production, a persistent challenge has emerged: explaining the bill. AWS addressed this directly with a new feature for Amazon Bedrock. It now supports cost allocation by IAM user and role, allowing companies to tag usage by team or department. This data feeds directly into AWS Cost Explorer, giving financial and technical leaders a clear view of inference spending across different models and projects.

Beyond cost management, AWS unveiled several other significant updates. Anthropic's most advanced model, Claude Mythos, entered a gated preview on Bedrock. Designed for cybersecurity, it can analyze vast codebases to identify complex vulnerabilities before they are exploited. Access is currently restricted to a select group of critical infrastructure organizations and open-source maintainers.

For businesses building with AI agents, AWS introduced the Agent Registry in preview. This private catalog, part of Amazon Bedrock AgentCore, lets teams register, discover, and govern AI agents, tools, and skills. The goal is to prevent redundant work and maintain oversight as agent deployments multiply.

Other notable releases include Amazon S3 Files, which presents S3 buckets as low-latency file systems, and updates to Amazon OpenSearch Service that unify metrics, logs, and AI agent traces. Amazon Braket also expanded quantum computing access by adding Rigetti's 108-qubit Cepheus processor. These updates reflect a clear trend: AWS is building out the foundational controls and specialized capabilities required for enterprise-scale AI operations.

Source: AWS