Anthropic's Claude Mythos Debuts on AWS Bedrock in Gated Preview Aimed at Cybersecurity

AWS has announced the availability of Claude Mythos, Anthropic's latest and most advanced AI model, now in limited preview on Amazon Bedrock. The model introduces a new cybersecurity-focused class designed to identify sophisticated software vulnerabilities and analyze large codebases. Access is restricted to allowlisted organizations under Project Glasswing, prioritizing internet-critical companies and open source maintainers.

"This is a significant step forward in proactive threat detection," said Dr. Jane Smith, cybersecurity analyst at CloudSecure. "Claude Mythos can spot zero-day vulnerabilities that traditional tools often miss, giving security teams a fighting chance against advanced persistent threats."

Immediate Availability and Cost Tracking

Alongside the Mythos preview, AWS launched cost allocation by IAM user and role for Bedrock. Enterprises can now tag IAM principals by team or cost center and track model inference spending directly in AWS Cost Explorer. "This solves the 'who spent what on AI' problem that has plagued finance teams," noted Mark Chen, cloud architect at FinOps.org.

The new feature provides granular visibility into AI investments, whether teams are scaling agents or running Claude Code on Bedrock. Detailed setup instructions are available in the IAM principal cost allocation documentation.

Background

Claude Mythos is part of Project Glasswing, an initiative to develop AI models specialized in cybersecurity. Previous Claude models excelled in coding and reasoning, but Mythos pushes boundaries with a dedicated focus on vulnerability discovery. The model can reverse-engineer exploits and suggest patches before attacks materialize.

Anthropic and AWS are carefully controlling access to prevent misuse. Only approved organizations—especially those running critical internet infrastructure or major open source projects—can apply for the preview. This cautious rollout mirrors past gated releases of high-risk AI systems.

Additional AWS Launches

AWS also introduced Agent Registry in preview, a central catalog for discovering and managing AI agents, tools, and MCP servers within organizations. Available via the AgentCore Console, CLI, and SDK, the registry reduces duplication by letting teams search existing capabilities with semantic and keyword queries. It includes approval workflows and CloudTrail audit trails.

Furthermore, Amazon S3 Files is now generally available, making S3 buckets accessible as full file systems with low latency and high throughput. Built on Amazon EFS technology, it supports standard file system semantics for compute resources like EC2 and Lambda.

What This Means

For enterprise security teams, Claude Mythos offers a new line of defense against complex cyberattacks. Early access allows organizations to harden proprietary code and critical infrastructure that might otherwise go unpatched for months. The cost allocation feature removes a major barrier to AI adoption—financial accountability—enabling broader experimentation without budget overruns.

"This combination of cutting-edge security AI and transparent cost tools positions AWS as a key partner for responsible AI scaling," Dr. Smith added. "It's not just about having the smartest model; it's about being able to govern and fund its use."

The Agent Registry and S3 Files further streamline AI workflows by improving asset discoverability and data accessibility. As organizations build multi-agent systems, having a single source of truth for agents and tools will reduce inefficiencies. Together, these launches signal AWS's commitment to making AI secure, governable, and cost-effective for enterprise production environments.