Saturday, May 9, 2026
Daily picks
20
articles scored
#1 GOLDReleaseClaude Code Releases
v2.1.136
- New `settings.autoMode.hard_deny` lets enterprises define classifier rules that ALWAYS block certain auto mode actions — no exceptions, no matter how Claude interprets user intent
- MCP OAuth refresh tokens now survive concurrent server refreshes — if you run multiple remote MCP servers, you should stop seeing daily re-authentication prompts
- Fixed: plan mode was supposed to block file writes, but an existing `Edit(...)` allow rule bypassed the block — this is now corrected
- Fixed: MCP servers in `.mcp.json`, plugins, and claude.ai connectors silently vanished after `/clear` in VS Code, JetBrains, and the Agent SDK
- WSL2 users: pasting images from the Windows clipboard now works via a PowerShell fallback when xclip/wl-paste can't read image data
#2 SILVERAnnouncementClaude Blog
Collaborate with Claude across Excel, PowerPoint, Word and Outlook
- Claude is now available directly inside Microsoft Office apps — Excel, PowerPoint, Word, and Outlook
- You can work with Claude on spreadsheets, presentations, documents, and email without leaving the Office suite
- Official Anthropic blog post confirms the integration is live for supported users
#3 BRONZEAnnouncementAnthropic Research
AlignmentMay 8, 2026Teaching Claude whyNew research on how we've reduced agentic misalignment.
- New Anthropic alignment research describes training Claude to understand *why* certain behaviors are unsafe — not just pattern-matching on surface rules, but internalizing the reasoning behind constraints
- The goal: an agent that generalizes correctly to novel situations it hasn't seen in training, rather than finding loopholes or failing in edge cases