Anthropic’s biggest story this week isn’t a product launch — it’s a federal lawsuit. After the U.S. Department of War designated Anthropic a national security supply chain risk, CEO Dario Amodei confirmed the company will challenge the ruling in court, alleging unlawful retaliation and First Amendment violation. In parallel, Anthropic shipped three significant enterprise features: the Claude Marketplace, Code Review for AI-generated PRs, and cross-platform memory import — all while the Claude Code CLI hit version 2.1.72 with voice mode and direct file writing.

Department of War Dispute: Anthropic Fights Back

The U.S. Department of War designated Anthropic a national security supply chain risk — language typically reserved for foreign adversaries — after contract negotiations broke down. The core dispute: the Pentagon wanted broader “lawful” military flexibility from Claude, including use cases Anthropic declined to support: fully autonomous weapons and mass domestic surveillance.

Secretary of War Pete Hegseth made the designation. President Trump subsequently banned federal agencies from using Anthropic technology and gave the DoD six months to transition. A $200 million Pentagon contract collapsed; OpenAI moved quickly to fill the gap.

On March 5, Dario Amodei published “Where Things Stand with the Department of War”, clarifying that the designation covers direct DoW contract use only — not Anthropic’s broader customer base — and that the company will continue supplying models to the DoD at nominal cost through the transition period. Anthropic is now challenging the designation in court.

Why it matters: This is historically unprecedented for a U.S. tech company. The outcome will set precedent for how AI safety constraints interact with federal procurement — and whether companies can hold the line on use-case restrictions without losing government contracts entirely. For enterprise customers, the designation has no effect on their Anthropic access today, but it’s a signal worth tracking.

Claude Marketplace: Enterprise Procurement, No Commission

Anthropic launched the Claude Marketplace on March 6 — an enterprise storefront where customers can purchase Claude-powered partner software against existing Anthropic spending commitments. Anthropic takes no commission on transactions.

Launch partners include Snowflake, GitLab, Harvey AI, Rogo, Replit, and Lovable Labs, spanning data management, development, legal automation, financial analysis, and app building. Snowflake’s inclusion is notable: it brings 12,600 customers via a $200M multi-year deal signed earlier in 2026.

Why it matters: The Marketplace accelerates Anthropic’s shift from pure API provider to platform. For enterprise buyers, it simplifies AI vendor consolidation — one spending commitment, multiple applications. For Anthropic, it deepens customer lock-in without the margin hit of a traditional app store. Watch for catalog expansion beyond the six launch partners; Anthropic has signaled datasets and additional services are coming.

Code Review: Multi-Agent PR Analysis Now in Research Preview

Claude Code Review launched in research preview for Teams and Enterprise customers on approximately March 9. The system dispatches multiple parallel agents to examine PRs for logic errors, security flaws, performance issues, and architectural concerns, then aggregates findings by priority and posts them directly to GitHub.

The internal Anthropic metric worth noting: the percentage of PRs receiving substantive review jumped from 16% to 54% after deployment. The <1% false positive rate — if it holds in production — would address the primary objection teams have to automated review tooling.

Pricing averages $15–25 per review, billed on token usage. Community reaction has been mixed: critics note the cost rivals unlimited subscriptions to traditional tools; proponents argue the multi-agent depth justifies it at enterprise scale.

Why it matters: Code Review represents Anthropic building out the full agentic development lifecycle — not just code generation, but systematic review of AI-generated output. For high-volume AI-assisted teams at companies like Uber, Salesforce, and Accenture, review throughput is the bottleneck. That said, teams should model the math carefully at their PR velocity before enabling it broadly.

Memory Import: Switching Platforms Just Got Easier

On March 2, Anthropic rolled out memory from chat history to all Claude users, including the ability to import memories from ChatGPT, Gemini, and other AI assistants. Users switching platforms can carry over personalized context without starting from scratch.

Why it matters: This is a low-key but strategically significant move. Switching costs have historically kept users locked into whichever AI platform they started on. Memory portability reduces that friction — and makes Claude a more credible default for organizations evaluating a primary AI platform. Enterprise teams that have accumulated context in other assistants now have a cleaner migration path.

Claude Code CLI: Rapid-Fire Updates

The CLI shipped multiple versions this week with notable additions:

v2.1.72 (March 10):

  • Direct file writing from code blocks — critical for SSH environments where clipboard isn’t available
  • Effort level visibility in the task spinner: low ○ / medium ◐ / high ●
  • Restored model parameter on the Agent tool for per-invocation overrides
  • Expanded auto-approval allowlist for git diff, cargo, and other common tools
  • Fixed Escape key unresponsiveness after query cancellation

v2.1.71 (March ~8):

  • VS Code upgrades: Spark icon for sessions, full session tabs
  • 11 new auto-approved bash commands
  • API reliability fixes for third-party gateways and proxy detection
  • OAuth token refresh elimination

v2.1.69–70:

  • Claude API skill for SDK app building
  • 10 new voice STT languages (Turkish, Dutch, Swedish, Russian, and others)
  • Claude Opus 4.6 defaulting to Medium effort

Voice mode is rolling out gradually, starting at 5% of users — allowing developers to issue natural-language voice commands instead of typing. No GA date announced.

API: Automatic Caching GA, Model Retirements

Automatic caching is now generally available on the Messages API. A single cache_control field enables the feature; the system auto-caches the last cacheable block and advances the cache point as conversations grow. Manual breakpoint management is no longer required. Available on the Claude API and Azure AI Foundry (preview).

Model retirements to act on now:

  • Claude Sonnet 3.7 and Claude Haiku 3.5 are retired — requests now return errors. Upgrade to Claude Sonnet 4.6 and Claude Haiku 4.5.
  • Claude Haiku 3 deprecation announced; retirement scheduled for April 19, 2026. Teams on claude-3-haiku-20240307 have roughly five weeks to migrate.

Web search and code execution tools are now generally available — no beta header required. Code execution is free when paired with web search or web fetch.

MCP: Enterprise Adoption Accelerates

The Model Context Protocol continued expanding beyond developer tooling this week:

  • Google Cloud released a gRPC transport package for MCP, extending support to enterprise microservices beyond the default JSON-RPC over HTTP. OpenAI, Microsoft, and Mistral AI confirmed MCP compatibility.
  • Glean introduced enterprise MCP support for safe AI actions across project management and data warehouse tools.
  • Media companies began adopting MCP to connect AI with CMS systems for automated editorial repurposing, SEO auditing, and pre-editorial checks.
  • An arXiv study analyzing 3,282 real-world MCP bugs identified common failure patterns around server resource handling and stale metadata — a useful reliability signal for teams building production MCP integrations.

Enterprise Scale: 300,000 Business Customers

Anthropic now serves over 300,000 business customers, with 500+ spending more than $1M annually. The $100K+ run-rate cohort grew approximately 7× year-on-year. Deloitte’s Claude deployment to ~470,000 employees remains the largest enterprise implementation on record.

What to Watch

  • DoW Lawsuit Resolution: Anthropic’s legal challenge is the most consequential near-term event in the AI industry. The outcome will establish whether AI safety constraints are viable in federal contracting — or whether companies must choose between safety commitments and government business.
  • Claude Haiku 3 Retirement — April 19, 2026: Teams still on claude-3-haiku-20240307 have roughly five weeks to migrate to Claude Haiku 4.5 before requests error out. Don’t let this slip.
  • Code Review GA Timing: Currently research preview for Teams/Enterprise only. Watch for broader availability and whether the per-review pricing adjusts based on community feedback.
  • Claude Marketplace Catalog Expansion: Anthropic has signaled plans to add datasets and additional services beyond the six launch partners. Worth tracking for enterprise procurement teams building AI vendor strategies.
  • Opus 4.6 Fast Mode Waitlist: Promises up to 2.5× faster output token generation. Waitlist open at claude.com/fast-mode; no GA date announced.

Check back next week for the latest from the Claude ecosystem. If you’re evaluating how these changes affect your enterprise AI strategy, Big Hat Group can help.