Claude Code Daily Briefing - 2026-03-21

Release Summary

Full Release Notes

New Features & Practical Usage

Claude Code v2.1.81 — Scripted Execution and Mobile Permission Relay (3/20)

Released the day after v2.1.80, v2.1.81 focuses on automation scenarios and remote permission management.

Key changes:

• --bare flag — Skips hooks, LSP, plugin sync, and skill directory walks for lightweight execution. Requires ANTHROPIC_API_KEY or apiKeyHelper via --settings. OAuth/keychain auth and auto-memory disabled. Ideal for scripted/automated calls
• --channels permission relay — Channel servers declaring the permission capability can forward tool approval prompts to your phone, enabling mobile-based permission management for remote sessions
• OAuth concurrent session fix — Multiple simultaneous Claude Code sessions no longer require re-authentication when one session refreshes its OAuth token
• Voice mode stabilization — Fixed silent retry failure swallowing with misleading “check your network” message; fixed audio not recovering when server silently drops WebSocket connection
• CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS fix — Now properly suppresses structured-outputs beta header, preventing 400 errors on Vertex/Bedrock proxy gateways
• Node.js 18 crash fix — Resolved crash on Node.js 18 environments
• Bash permission prompt fix — Eliminated unnecessary permission prompts for Bash commands containing dashes in strings
• Worktree session restore — Resuming a session in a worktree now automatically switches back to that worktree
• MCP tool call display — MCP read/search tool calls collapse into a single “Queried {server}” line (expand with Ctrl+O)
• ! bash mode discoverability — Claude now suggests ! mode when running interactive commands
• Plugin freshness — Ref-tracked plugins now re-clone on every load to pick up upstream changes
• Remote Control improvements — Auto-generated titles, /rename sync fix, /exit archive stabilization
• MCP OAuth CIMD support — Client ID Metadata Document (SEP-991) support for servers without Dynamic Client Registration
• Plan mode change — “Clear context” option hidden by default (restore with "showClearContextOnPlanAccept": true)
• Windows line-by-line streaming disabled — Disabled on Windows (including WSL) due to rendering issues

GitHub Releases

Court Filing Reveals Pentagon Told Anthropic Sides Were “Nearly Aligned” (3/20)

Anthropic submitted two sworn declarations to a California federal court, directly contradicting the Pentagon’s “national security threat” claims.

Key revelations:

• “Very close” email: On March 4 — the day after the Pentagon formally finalized its supply-chain risk designation — Under Secretary Emil Michael emailed Amodei saying the two sides were “very close” on the exact two issues the government now cites as evidence of a national security threat: autonomous weapons and mass surveillance
• The contradiction: The Pentagon is now claiming the very same issues make Anthropic a national security threat, despite its own official saying they were nearly aligned just a week after Trump declared the relationship over
• Sarah Heck (Head of Policy): Filed sworn declaration on Anthropic’s behalf
• Thiyagu Ramasamy (Head of Public Sector): Filed sworn declaration on technical details
• 3/24 hearing D-3: Preliminary injunction hearing before Judge Rita Lin in 3 days

TechCrunch

Anthropic Officially Denies “Backdoor or Kill Switch” (3/21)

Anthropic’s Head of Public Sector Thiyagu Ramasamy officially denied the ability to sabotage military AI tools in a court filing.

Key points:

• “Never had the ability”: “Anthropic has never had the ability to cause Claude to stop working, alter its functionality, shut off access, or otherwise influence or imperil military operations”
• No backdoor or kill switch: Does not maintain any remote ‘kill switch’ or backdoor; personnel cannot log into DoD systems to modify or disable models
• Update process: Claude updates require approval from the government and its cloud provider (AWS)
• 3/24 hearing D-3: Pre-emptive move to defuse sabotage concerns ahead of the hearing

National Today

Developer Workflow Tips

v2.1.81 --bare — Lightweight Claude Code for Automation

Use --bare in CI/CD pipelines and automation scripts to skip all unnecessary initialization overhead.

ANTHROPIC_API_KEY=sk-... claude -p "Review this code" --bare

Skips hooks, LSP, plugin sync, and skill directory walks for fast, clean execution.

Mobile Permission Relay with --channels

Run Claude Code on a remote server and approve tool permissions from your phone. Channel servers with the permission capability forward approval prompts to your mobile device.

Collapsed MCP Tool Calls

In v2.1.81, MCP read/search tool calls now collapse into a single line for cleaner output. Press Ctrl+O to expand details.

Security & Limitations

Claude Opus & Sonnet 4.6 Service Incident (3/21)

Elevated error rates hit Claude Opus and Sonnet 4.6 on March 21.

Incident summary:

• Scope: Primarily affected free-tier users worldwide; over 5,000 complaints at peak
• Symptoms: API Error 500, chat and coding tools unavailable
• Timeline: Opus 4.6 recovered first, followed by Sonnet 4.6
• Status: Resolved

Claude Status

Anthropic vs. Pentagon — 3/24 Hearing D-3, “Nearly Aligned” Bombshell (3/21)

With the 3/24 hearing three days away, Anthropic has mounted a strong counterattack.

Latest developments:

• “Nearly aligned” email revealed: Pentagon Under Secretary emailed Amodei saying sides were “very close” on the two disputed issues — the day after supply-chain risk designation was finalized
• Sabotage ability officially denied: Anthropic’s Head of Public Sector declared under oath that no backdoor, kill switch, or remote access to DoD systems exists
• Foreign workforce security concerns persist: Pentagon’s claims about Chinese national employees continue as part of the legal battle
• 3/24 Judge Rita Lin: San Francisco federal court to decide whether to pause the supply-chain risk designation
• Revenue at risk: Anthropic CFO warned of “hundreds of millions” in potential 2026 revenue losses

TechCrunch | Axios

Ecosystem & Plugins

Claude Code Channels Trending on Hacker News (3/20–21)

The Claude Code Channels feature introduced as a research preview in v2.1.80 is trending at #4 on Hacker News with 341 points.

• Core concept: Transforms Claude Code from a request-response tool into an event-driven autonomous agent
• Use cases: Receive CI build completions, monitoring alerts, and Discord/Telegram messages directly in your session
• v2.1.81 addition: --channels permission relay enables mobile tool approval

VentureBeat

Claude Double Usage Promotion Ongoing (3/13–3/27)

Anthropic is running a usage limit doubling promotion through March 27.

• When: Weekday off-peak hours (outside 8AM–2PM ET), weekends 24/7
• Who: Free and Pro plan users
• 6 days remaining in the promotion

Android Headlines

Community News

• Claude Code v2.1.81 released: --bare lightweight mode, --channels mobile permission relay, OAuth concurrent session fix, voice mode stabilization. GitHub

• Anthropic reveals “nearly aligned” email: Pentagon Under Secretary told Amodei sides were “very close” the day after supply-chain risk designation. TechCrunch

• Anthropic denies sabotage capability: No backdoor, kill switch, or DoD system access declared under oath. National Today

• Claude Opus & Sonnet 4.6 service incident: Elevated error rates resolved. Claude Status

• Claude Code Channels trending #4 on HN: Event-driven agent concept generating significant community interest. VentureBeat

• Claude 2x promotion — 6 days left: Off-peak doubled limits through 3/27. Android Headlines

Minor Changes

• v2.1.81 plan mode “clear context” hidden: Clear context option after plan acceptance now hidden by default. Restore with "showClearContextOnPlanAccept": true.
• v2.1.81 Windows line-by-line streaming disabled: Disabled on Windows and WSL due to rendering issues.
• v2.1.81 Remote Control session titles: Auto-generated from first prompt instead of generic titles. Refreshes after third message.
• v2.1.81 tmux clipboard fix: Fast Cmd+Tab followed by paste no longer loses clipboard content.
• v2.1.81 /btw fix: Pasted text now properly included when used during active response.
• v2.1.81 invisible hook attachments: No longer inflate message count in transcript mode.

Recommended Reads

• “New court filing reveals Pentagon told Anthropic the two sides were nearly aligned — a week after Trump declared the relationship kaput”: TechCrunch details the contradiction exposed by Anthropic’s sworn declarations. TechCrunch

• “Anthropic Denies Ability to Sabotage Military AI Tools”: Analysis of Anthropic’s official denial of backdoor and kill switch capabilities in military deployments. National Today

• “Anthropic just shipped an OpenClaw killer called Claude Code Channels”: VentureBeat examines how Channels transforms Claude Code into an event-driven autonomous agent. VentureBeat

Tomorrow’s Watch List

• Anthropic vs. Pentagon 3/24 hearing D-2: Watch for Pentagon’s response to the “nearly aligned” email revelation and sabotage denial. Judge Rita Lin may issue pre-hearing questions.
• v2.1.81 --bare adoption: CI/CD pipeline and automation script usage patterns expected to emerge.
• Claude Channels community momentum: Continued HN visibility likely to drive MCP server implementations and integration examples.
• Claude 2x promotion — 6 days left: Community strategies for maximizing doubled limits before 3/27 end.
• Service stability monitoring: Post-incident observation following today’s Opus/Sonnet elevated error rates.