Anthropic AI Updates: April 8, 2026

1. Claude Code v2.1.94 Adds Bedrock Mantle Support and Raises Default Effort Level

Anthropic. Claude Code v2.1.94, released April 7, ships several developer-facing additions. The release adds support for Amazon Bedrock powered by Mantle (enable with CLAUDE_CODE_USE_MANTLE=1) and raises the default effort level from medium to high for API-key, Bedrock, Vertex, Foundry, Team, and Enterprise users — the effort level remains adjustable via /effort. Additional additions include a compact Slacked #channel header with a clickable link for Slack MCP send-message tool calls, a keep-coding-instructions frontmatter field for plugin output styles, and hookSpecificOutput.sessionTitle support on UserPromptSubmit hooks for setting the session title. Plugin skills declared via "skills": ["./"] now use the skill’s frontmatter name field for invocation, providing stable naming across installation methods. Source

2. Claude Messages API Now Available Directly on Amazon Bedrock

Anthropic. The Anthropic Messages API is now available on Amazon Bedrock as a research preview via a new /anthropic/v1/messages endpoint that uses the identical request shape as the first-party Claude API. The endpoint runs on AWS-managed infrastructure with zero operator access, currently available in us-east-1. This allows teams using Bedrock to switch between the Anthropic API and Bedrock without code changes — a significant operational simplification for enterprises that need AWS billing and compliance while also targeting first-party API parity. Access is by request through an Anthropic account executive. Source

3. Anthropic Launches Project Glasswing to Deploy AI Defensively Against Zero-Day Vulnerabilities

Anthropic. Anthropic has announced Project Glasswing, a major collaborative cybersecurity initiative alongside AWS, Apple, Cisco, Google, Microsoft, NVIDIA, and the Linux Foundation. The project deploys Claude Mythos Preview — an unreleased frontier model — to identify zero-day vulnerabilities in critical software at scale. According to Anthropic, the model has already discovered thousands of high-severity flaws in major operating systems and web browsers, with findings shared across more than 40 organizations that maintain critical infrastructure.

To support the initiative, Anthropic is committing $100 million in model usage credits for participating organizations and $4 million in direct donations to open-source security groups, with plans to publish learnings and best practices within 90 days. Glasswing reflects Anthropic’s view that while AI-powered vulnerability discovery poses real risks, deploying these capabilities defensively — and first — can dramatically benefit the broader security ecosystem before malicious actors do the same. Source