Anthropic Launches Claude Opus 5 on AWS Bedrock: Zero Data Retention by Default

Anthropic and AWS have announced the general availability of Claude Opus 5 on Amazon Bedrock and the Claude Platform on AWS, marking the debut of Anthropic’s fifth-generation model family.

Released on July 24, 2026, Opus 5 arrives as Anthropic’s most capable model to date, targeting enterprise workflows in agentic coding, long-running autonomous tasks, and document-heavy analysis, all while maintaining zero data retention (ZDR) by default on Bedrock.

For security teams evaluating AI adoption, this launch is notable less for raw benchmark gains and more for how Anthropic and AWS are packaging enterprise trust controls directly into the deployment model.

Anthropic Launches Claude Opus 5 on AWS Bedrock

According to Anthropic, Opus 5 reportedly matches the top-tier intelligence of Claude Fable 5 across many domains, but at Opus-tier pricing, a positioning move aimed at cost-conscious enterprise buyers. The model claims improvements in three areas relevant to security and engineering teams:

Codebase navigation and production-quality code generation with adaptive strategy shifts mid-task
Long-running agentic reliability, including error recovery and obstacle navigation across multi-hour or overnight jobs
Deeper reasoning over long documents, with the largest accuracy gains on compliance-heavy enterprise tasks

Anthropic also states Opus 5 improves on Claude Opus 4.8’s cybersecurity-relevant capabilities across coding and security domains. Notably, in higher-risk scenarios, the model can fall back to Opus 4.8, with users notified and fallback behavior configurable via the API a guardrail worth testing for teams building autonomous security tooling.

The Bedrock integration is built on AWS’s next-generation inference engine, offering zero operator access and regional data residency alongside ZDR by default. This distinguishes it from typical third-party model integrations where data handling terms require separate negotiation.

The Claude Platform on AWS option supports ZDR on request and mirrors Anthropic’s native console experience while unifying billing and authentication under AWS. For threat intelligence teams already operating within AWS environments, this reduces the attack surface associated with managing separate vendor credentials and API keys.

Opus 5 is accessible via the Bedrock console Playground, the Anthropic Messages API through bedrock-runtime, or the standard Invoke and Converse APIs via AWS CLI and SDKs.

Prerequisites include an active Bedrock-enabled AWS account, Boto3, the anthropic[bedrock] SDK, and IAM permissions covering bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream, and bedrock:CreateInference.

A notable protocol change: Opus 5 supports adding or removing tools mid-conversation using tool_addition and tool_removal content blocks on system-role messages, eliminating the need to resend the full tools array on every turn a meaningful efficiency gain for teams building multi-tool security automation agents.

Claude Opus 5 is live today on Bedrock in US East (N. Virginia), Asia Pacific (Melbourne), Europe (Ireland), and Europe (Stockholm), with broader regional coverage detailed in AWS documentation. Claude Platform on AWS extends availability across North America, South America, Europe, and Asia Pacific.

Teams looking to optimize existing prompts for Opus 5 can use Bedrock’s Advanced Prompt Optimization tool, which benchmarks current prompts against evaluation criteria and generates production-ready rewrites.