Claude Design Ships — Anthropic Steps Directly Into Figma’s Lane
Anthropic Labs launched Claude Design on Friday morning, a conversational design product that turns prompts into interactive prototypes, presentation decks, one-pagers, wireframes, and marketing assets. The backend is Opus 4.7, the same vision model that went GA two days earlier; users describe what they want and Claude Design generates the artifact, including code-powered prototypes with voice, video, 3D, and AI elements baked in. A key differentiator Anthropic is pushing hard: Claude Design can read a team’s codebase and existing design files to apply the company’s design system automatically to every new asset.
It’s in research preview today for Pro, Max, Team, and Enterprise subscribers, rolling out across the day. Canva is the named launch partner — teams can finish a deck or prototype in Claude Design, export, and drop it straight into Canva for collaborative editing. The market took it seriously: Figma closed down about 7%, Adobe down about 4% on the news. The context nobody’s missing — CPO Mike Krieger quietly resigned from Figma’s board on April 14, the same day The Information reported Opus 4.7 would include design capabilities.
Amodei at the White House: First Clear Sign the Pentagon Freeze Is Cracking
Dario Amodei walked into the West Wing Friday afternoon for a closed-door meeting with White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent. Axios, CNN, and PBS all put the meeting in the same frame: a thaw. The Trump administration had severed ties with Anthropic after the company refused to allow Claude for “all lawful purposes,” including autonomous weapons and mass surveillance — a posture the Pentagon escalated by labeling Anthropic a “supply-chain risk,” a tag historically reserved for foreign-adversary-linked vendors. Anthropic sued; a California federal judge blocked the label last month.
What shifted: Mythos. Bessent and Fed Chair Powell already briefed bank CEOs on April 7 about Mythos’s cyber risk. CISA and parts of the intelligence community are now testing it. Treasury has asked for access. The White House meeting was described as “productive” by CNN’s sources; nobody on either side is calling it resolved. Read: a federal customer lane for Mythos is being quietly negotiated alongside the public safety-policy fight.
Cyber Verification Program Opens — Approved Security Researchers Get Unrestricted Opus 4.7
Anthropic opened applications for its new Cyber Verification Program (CVP) alongside the Opus 4.7 release. The deal: Opus 4.7 ships with automatic real-time classifiers that detect and block requests tied to prohibited or high-risk cyber activity, which will trip up legitimate vulnerability researchers, penetration testers, and red-teamers doing their jobs. The CVP is a free, application-based carve-out that verifies identity and grants those professionals access to the model without the default refusals.
The framing is important: Anthropic is explicitly positioning CVP as the proving ground for eventually broadening access to Mythos-class models. In other words — pass CVP, and you’re building the trust infrastructure that lets Mythos ship more widely later. This is how Anthropic thinks “responsible deployment” scales.
Claude Code Ships Auto Mode for Max, Plus /ultrareview and /effort Controls
The Claude Code 2.1.101 train that landed alongside Opus 4.7 adds three things worth knowing about. Auto mode is now open to Max subscribers: long-horizon runs — deep research, refactors, feature builds that iterate until a benchmark passes — execute without asking for confirmation at each step. /ultrareview is a new slash command that spawns a dedicated multi-agent code review session in the cloud, doing parallel passes for bugs, edge cases, security issues, and design problems. Pro and Max users get three free ultrareviews to try it. /effort exposes the effort-level dial directly in the REPL so devs can flip between medium, high, and xhigh per session without restarting.
Also in the release: a new terminal-matching theme, noticeably fewer permission prompts, and a batch of Windows fixes. Taken together with the model drop, Anthropic has now shipped Routines, Code Review, Advisor, Monitor Tool, Managed Agents, Auto mode, and /ultrareview within eight days. The harness roadmap is moving faster than the model roadmap.
MCP Connector Roundup: Zoom, Microsoft 365, Sentinel, and Opera All Live This Week
Four new or newly-expanded MCP connectors for Claude landed this week and are worth putting on the integration radar. Zoom’s MCP server brings meeting summaries, transcripts, recordings, and scheduling into Claude Cowork and Claude Code — query history in natural language, generate follow-ups without leaving Claude. Microsoft 365 (Outlook, SharePoint, OneDrive, Teams) is now available to all Claude plans including free, not just enterprise; access runs through user-delegated permissions via the Microsoft app marketplace. Microsoft Sentinel shipped a third-party MCP connector that lets Claude pull SIEM data while keeping Entra ID for auth. Opera’s Browser Connector ties Claude directly to the browser for in-page context.
The volume is the story. Zoom, Microsoft, Opera, Flexibits, Canva, GitHub, and dozens of smaller vendors have all shipped MCP endpoints in the last 60 days. MCP is on track to be the default protocol for LLM-app integration, the way REST became the default for web APIs.
One Day Left: Claude Haiku 3 Retires Tomorrow, April 19
Tomorrow is the hard cutoff for claude-3-haiku-20240307. After Sunday, API calls against that model string will fail outright — not warnings, not soft fallbacks, errors. Drop-in migration is Claude Haiku 4.5 (claude-haiku-4-5-20251001): same price tier, materially faster, better eval scores across the board. If you have cron jobs, ETL pipelines, RAG systems, or legacy agents still pinned to the old string, today is the last day to ship the change and catch issues in staging.
Downstream dates on the calendar: the 1M-token context beta for Sonnet 4.5 and Sonnet 4 expires April 30; Opus 4 and Sonnet 4 both fully retire June 15. Migration targets are Opus 4.7 and Sonnet 4.6.
OpenAI Ships GPT-5.4-Cyber — The Anti-Mythos Playbook
OpenAI dropped GPT-5.4-Cyber on April 16, one day after Anthropic put Opus 4.7 out and a week after the Mythos Preview “consortium” announcement. It’s GPT-5.4 with refusal layers deliberately relaxed for defensive security work — the same model family, but with binary reverse engineering and malware analysis unlocked for verified researchers. Access is gated through OpenAI’s Trusted Access for Cyber (TAC) program, but the target is thousands of individual defenders and hundreds of security teams — two orders of magnitude wider than Anthropic’s 40-partner Mythos list.
The framing is a direct shot: OpenAI publicly said it “doesn’t think it’s practical or appropriate to centrally decide who gets to defend themselves,” which is as close to naming Anthropic’s consortium model as you get in vendor diplomacy. Two different bets on the same problem — concentrated responsibility (Anthropic) vs. broad but verified access (OpenAI). Enterprises running red teams now have a real choice. Watch procurement move.
Investor Offers Push Anthropic Valuation Toward $800B — IPO Talks Heat Up
Anthropic closed its Series G at a $380B post-money in February. Two months later, The Next Web and TechFundingNews report that new investor offers are landing at roughly twice that — around $800B — and the company is in early talks with Goldman Sachs, JPMorgan, and Morgan Stanley about a potential IPO as soon as October. The underlying number that’s driving all of this is revenue: from $1B annualized at end-of-2024 to $30B+ by early April 2026. Anthropic is now generating more revenue than OpenAI on a run-rate basis for the first time, per TradingKey.
The employees are reading the tea leaves. Anthropic’s most recent $350B tender offer for private shares reportedly fell short of its target — staff are sitting on equity rather than selling, betting the IPO prices higher. If it lists in Q4 at $600B+ the way the banks are reportedly modeling, it’s the largest tech IPO since Saudi Aramco and the biggest in U.S. history.
Anthropic Is No Longer a Model Lab. It’s a Platform Company Now.
String together the last eight days and the shape is unmistakable. Routines. Opus 4.7. Advisor. Monitor. Managed Agents. Code Review. /ultrareview. Auto mode. Claude Design. Each one ships a layer of the stack above the model — orchestration, observability, sandboxing, review, UI generation. The $20/seat unbundling from last week is the pricing signal that ties it together: flat-fee seats are how you sell access to a chat product; metered compute-on-your-behalf is how you sell agentic infrastructure. Anthropic isn’t raising prices. It’s repricing to match the product it’s actually becoming — an AWS for autonomous work.
Claude Design is the public tell. A lab doesn’t ship a Figma competitor; a platform company does. Mike Krieger — the Instagram co-founder hired as CPO specifically to turn Anthropic into a product company — quietly leaves Figma’s board on Tuesday and ships a Figma competitor on Friday. That’s a roadmap statement. The question for customers has stopped being “which model is best?” and become “whose harness and application layer do we want to build on?” The IPO talk at $800B isn’t priced against the model. It’s priced against the platform.