Pentagon Appeals Court Ruling That Blocked Anthropic’s National Security Designation
The Department of War filed an appeal on April 2 challenging a federal judge’s preliminary injunction that temporarily blocked the Pentagon’s supply-chain risk designation against Anthropic. U.S. District Judge Rita F. Lin in San Francisco had ruled the designation likely violated constitutional protections, allowing Anthropic to continue working with federal agencies while the case proceeds. The Pentagon is arguing the designation was lawful and falls within its authority to protect military procurement chains.
The original designation — the first known application of the supply-chain risk law to a US-based company — came after Anthropic refused to remove safety guardrails that would enable fully autonomous weapons and mass surveillance. If the appeal succeeds, Anthropic could be cut off from doing business with federal agencies, contractors, and suppliers. The case is being closely watched as a precedent-setting test of how far the government can reach into AI company policy decisions.
UK Courts Anthropic with London Expansion Pitch and Dual Stock Listing Offer
Britain is actively pitching Anthropic on a London office expansion and potential dual listing on the London Stock Exchange, positioning itself as a more favorable regulatory environment than either Washington or Brussels. Prime Minister Keir Starmer’s office is backing the effort, with proposals set to be presented to CEO Dario Amodei during a late-May visit. The UK government’s pitch is explicitly tied to Anthropic’s US defense clash — framing Britain as a place where AI companies can operate without being weaponized by military procurement battles.
The pitch has a tangible foundation: Anthropic already partners with the Department for Science, Innovation and Technology (DSIT) on safe AI deployment in government services. The GOV.UK AI assistant — powered by Claude and providing career guidance to citizens — is one of the first major outcomes of that collaboration. For Anthropic, the UK relationship offers strategic diversification at a moment when its US government standing is in legal limbo.
Models API Gets Capability Fields; Message Batches Hit 300k Token Cap
Anthropic pushed a batch of API improvements this week. The Models API (GET /v1/models and GET /v1/models/{model_id}) now returns max_input_tokens, max_tokens, and a capabilities object, making it easier to programmatically select the right model for a task. The Message Batches API now supports a 300k max token cap for Claude Opus 4.6 and Sonnet 4.6, available via the output-300k-2026-03-24 beta header — a major unlock for long-form content, structured data extraction, and large code generation tasks.
Two more quality-of-life improvements also landed: automatic caching now works by adding a single cache_control field to request bodies (no manual breakpoint management required), and data residency controls let developers specify where model inference runs via the inference_geo parameter. US-only inference is available at 1.1x pricing for models released after February 1, 2026 — a meaningful addition for regulated industries that need data to stay domestic.
Claude Sonnet 3.7 and Haiku 3.5 Retired; 1M Token Beta Ends April 30
Anthropic retired Claude Sonnet 3.7 and Claude Haiku 3.5, pushing developers to upgrade to Claude Sonnet 4.6 and Claude Haiku 4.5 respectively. The retirements come as the 4.x generation consolidates its dominance across benchmarks. Separately, the 1M token context window beta for Claude Sonnet 4.5 and Claude Sonnet 4 will end on April 30, 2026. After that date, developers who need 1M token context should migrate to Claude Sonnet 4.6 or Claude Opus 4.6, which both support the full 1M token window at standard pricing with no beta header required.
The retirement of the beta header requirement for 1M context is a meaningful simplification for developers who had to track and manage experimental flags. If you’re running any production workflows against Sonnet 3.7 or the 1M beta, now is the time to migrate — April 30 is three and a half weeks away.
Claude Code Adds Bedrock Setup Wizard, Cost Breakdown, and Interactive Release Notes
The latest Claude Code releases shipped several developer-facing improvements. An interactive Bedrock setup wizard is now accessible from the login screen, simplifying enterprise AWS deployments. The /cost command now includes per-model and cache-hit breakdowns for subscription users, making it easier to track where compute is actually going in a session. And /release-notes has been upgraded to an interactive version picker, letting developers jump to changelogs for any specific version without leaving the terminal.
A new policy setting, forceRemoteSettingsRefresh, blocks CLI startup until remote managed settings are freshly fetched — useful for enterprise teams that need to enforce policy changes before any session begins. The /powerup command, introduced April 1, continues to gain traction: it delivers interactive lessons teaching Claude Code features with animated demos directly in the terminal, making onboarding significantly more accessible for new users.
Claude Can Now Search Outlook and Teams for Free
Claude AI has expanded its Microsoft productivity integrations, with the ability to search Outlook email and Microsoft Teams conversations now available at no additional cost. The integration puts Claude directly inside the workflows where enterprise knowledge lives — enabling use cases like summarizing email threads, surfacing relevant messages before a meeting, and drafting responses informed by prior conversation context. It’s a natural pairing given that Claude is already deployed as the GOV.UK AI assistant and is embedded in a growing number of enterprise workflows.
For enterprise IT teams, the Outlook and Teams integration expands the surface area of Claude without requiring custom API work. Combined with Anthropic’s recent data residency controls, organizations in regulated industries can now start mapping out a Claude deployment that keeps inference domestic and integrates with the tools their teams already use daily.
Claude Mythos Timeline: Polymarket at 40% for Q2, April Still on the Table
As the legal and geopolitical noise around Anthropic builds, the developer community is keeping a close eye on Claude Mythos — the leaked frontier model that Anthropic describes as a “step change in capabilities.” Polymarket prediction markets currently put the odds of a Q2 2026 general release at roughly 40%. Separate analysis from WaveSpeed AI outlines what a Mythos API might look like at launch: pricing likely above Opus 4.6, access gated behind an API tier similar to how GPT-o1 launched, and extended context as a key differentiator.
The model remains expensive to run at scale and is available only to a small early-access group. But the leaked draft blog post and internal codename mapping (Capybara = new tier above Opus) suggest Anthropic is further along than typical pre-announcement silence would imply. An April release is still possible — but Q2 is the safer bet given the operational and legal headwinds the company is currently managing.
The Anthropic-Pentagon Battle Is a Preview of Every AI Company’s Future Dilemma
The Anthropic blacklisting saga is getting framed as a story about one company versus the US government. It’s actually a preview of the decision every major AI lab will face: how much control over your product’s use cases are you willing to trade for government access? Anthropic drew its line at autonomous weapons and mass surveillance. The Pentagon said that line made the company a supply-chain risk. A federal judge blocked the designation. The DoW appealed. Meanwhile, the UK is waiting with a friendlier pitch.
The downstream signal here matters: the US government wants AI tools it can deploy without safety guardrails getting in the way. If that becomes the standard condition for federal contracts, every AI company faces the same calculus Anthropic is navigating now. The UK’s play is sharp — not just economically, but as a regulatory positioning story. Britain is betting that “responsible AI partner” is a more durable brand than “unrestricted weapons tool.” For AI companies trying to serve both defense and civilian markets without compromising core safety commitments, the Anthropic case is the test run everyone is watching.