Anthropic’s frontier AI model, Claude Fable 5.1, is now available on AWS through Amazon Bedrock and the Claude Platform. This rollout anchors a major infrastructure update featuring the Amazon Linux 2027 public preview, Graviton5-powered EC2 instances, and expanded Apache Iceberg v3 support in Amazon Redshift.
Claude Fable 5.1 Deployment and Governance on AWS
Last week, Anthropic’s Claude Fable 5.1 landed on Amazon Bedrock and the Claude Platform on AWS. According to official disclosures, the model is built for long-running, high-stakes workloads that span multiple applications and execute over hours. It handles intricate tasks across an entire codebase, code review, and performance optimization independently.
Anthropic has classified Fable 5.1 as a “Covered Model.” This designation triggers strict data retention, safety review, and access policies. Deployments utilizing the model are subject to data retention for up to 30 days alongside human review by Amazon personnel via the newly introduced aws_review data retention mode. Within this specific mode, AWS retains prompts and outputs exclusively for human safety checks inside the secure AWS boundary.
By contrast, the provider_data_share mode is legacy, ensuring Amazon Bedrock does not share proprietary customer data with the model provider. Furthermore, Enterprise Frontier Safeguards (EFS)—developed jointly by AWS and Anthropic—allow eligible enterprises to leverage these Covered Models while retaining control over their cloud environment.
Infrastructure Upgrades: AL2027 and Graviton5 Instances
Underneath the AI layer, AWS rolled out significant hardware and operating system updates. Amazon Linux 2027 (AL2027) entered public preview. Running on kernel 7.1+, AL2027 targets cloud-native workloads requiring high scale, performance, and security. It serves web applications, databases, containerized microservices, and AI/ML workloads.
Simultaneously, AWS introduced Amazon EC2 R9g and R9gd memory-optimized instances. Powered by custom AWS Graviton5 processors, these instances deliver up to 25% better compute performance than their Graviton4-based predecessors (R8g and R8gd). AWS benchmarks show these chips are up to 30% faster for database operations, and up to 35% faster for both web applications and machine learning workloads.
Serverless and Data Lakehouse Enhancements
Developer velocity and data engineering also saw notable tooling expansions. AWS Lambda SnapStart, previously restricted to managed runtimes like Python, .NET, and Java, now supports container image functions. This opt-in capability slashes cold start latencies from several seconds down to sub-second thresholds for latency-sensitive APIs and ML inference tasks.
For data engineers, Amazon Redshift added read-and-write support for Apache Iceberg v3 tables. Provisioned and serverless clusters running on Graviton can now leverage v3 capabilities, including default column values, row lineage, and deletion vectors.
Meanwhile, the AWS Agent Registry reached general availability. It offers a private, governed catalog for discovering enterprise agents, tools, skills, MCP servers, and custom resources, backed by AWS CloudTrail audit logs and semantic search.
