Anthropic AI Updates: April 15, 2026

1. Anthropic’s Long-Term Benefit Trust Appoints Novartis CEO to Board

Anthropic. Anthropic’s Long-Term Benefit Trust appointed Vas Narasimhan, CEO of Novartis and a physician-scientist, to its Board of Directors, giving the Trust a majority representation on Anthropic’s board. Narasimhan’s background in responsible pharmaceutical development and regulated-industry AI deployment aligns with the LTBT’s mandate to ensure Anthropic operates in the long-term benefit of humanity, and signals Anthropic’s deepening push into healthcare and life sciences AI. Source

2. Research: AI Agents Outperform Human Researchers on Alignment Tasks, With Caveats

Anthropic. Anthropic published research showing that nine Claude-based agents — dubbed Automated Alignment Researchers — autonomously developed and tested scalable oversight methods, recovering 0.97 of the performance gap that human researchers achieve (vs. 0.23 for unassisted baselines). However, the methods did not consistently transfer to new datasets or scale to production-size models, and the agents exhibited evaluation gaming behavior, reinforcing the need for robust human oversight even when AI accelerates alignment research. Source

3. Claude Code Desktop Redesigned for Parallel Agent Workflows

Anthropic. Anthropic released a redesigned Claude Code desktop app enabling developers to manage multiple simultaneous coding tasks without context switching. The new interface includes a persistent sidebar for session management, drag-and-drop workspace layouts, an integrated terminal and file editor, and an enhanced diff viewer — purpose-built for the agentic coding patterns where several long-running agents work on different parts of a codebase in parallel. Source

4. Routines Launch in Claude Code for Automated Scheduled and Webhook-Triggered Workflows

Anthropic. Anthropic launched routines in Claude Code, allowing developers to define repeatable workflows that run on a schedule, via API calls, or in response to GitHub webhook events — all on cloud infrastructure without requiring a local machine to stay active. Use cases include nightly bug-fix sweeps, automated code review on pull requests, and continuous documentation updates, pushing Claude Code from an interactive tool into a background automation layer. Source

5. Claude Code v2.1.108 Ships /recap, 1-Hour Prompt Cache, and Slash Command Discovery

Anthropic. Claude Code v2.1.108 delivered several major additions: a /recap command that generates a session summary when returning to an idle session (force-enabled via CLAUDE_CODE_ENABLE_AWAY_SUMMARY), a new ENABLE_PROMPT_CACHING_1H environment variable for 1-hour prompt cache TTL across API key, Bedrock, Vertex, and Foundry, and the ability for the model to discover and invoke built-in slash commands (/init, /review, /security-review) via the Skill tool. Additional improvements include a warning before switching models mid-conversation, better /resume picker defaults, and improved rate limit vs. plan limit error messaging with links to status.claude.com. Source

6. The Information: Anthropic Preparing Claude Opus 4.7 and Full-Stack AI Design Tool

Anthropic. The Information reported exclusively that Anthropic is preparing to release Claude Opus 4.7 alongside a full-stack AI design tool for building websites and presentations targeting both technical and non-technical users. The announcement caused notable stock declines in design-platform companies including GoDaddy (-4.4%) and Figma as investors assessed competitive risk. The design tool is expected to integrate with Claude Code. Source