Anthropic had its most consequential product week of 2026. Claude Managed Agents launched in public beta alongside the new ant CLI, delivering a full cloud-hosted agent platform for the first time. Simultaneously, the company revealed Project Glasswing โ a restricted cybersecurity initiative powered by the unreleased Claude Mythos Preview model that tops 17 of 18 benchmarks โ and closed a ~$400M acquisition of Coefficient Bio. Revenue has crossed $30B annualized run rate, and an October 2026 IPO is no longer speculation. Here is everything that matters for enterprise teams this week in Claude Weekly.
Claude Managed Agents: The Platform Bet
The biggest announcement of the week โ arguably of the year โ is Claude Managed Agents, now in public beta via the Claude Developer Platform.
This is not another SDK wrapper. Managed Agents is a cloud-hosted agent execution layer: secure sandboxing, state management, built-in MCP toolsets (GitHub, Slack, Linear, and more), per-user credential vaults, webhooks for human-in-the-loop interrupts, and server-sent event streaming. Early adopters include Notion, Asana, and Sentry. Access requires the managed-agents-2026-04-01 beta header.
The companion ant CLI ships alongside it โ a new command-line client for the Claude API offering YAML-based versioning of API resources, native Claude Code integration, and streamlined scripting without the full agent loop.
For enterprise teams evaluating agentic architecture, this matters because it removes the two hardest infrastructure problems: credential delegation and execution isolation. Anthropic is now competing not just on model quality but on agent ops.
Claude Cowork GA: Enterprise Controls Land
Claude Cowork moved from research preview to general availability for all paid subscribers on April 9. The GA release adds the controls enterprise buyers need: role-based access controls, group spend limits, usage analytics, expanded OpenTelemetry support, a Zoom MCP connector, and per-tool connector controls.
Three months of preview gave Anthropic time to harden the product. Paired with the Claude Managed Agents beta, this represents a cohesive enterprise suite โ collaborative sessions up top, agentic infrastructure underneath.
Project Glasswing: The Model They Couldn’t Release
On April 7, Anthropic announced Project Glasswing and, with it, the existence of Claude Mythos Preview โ a frontier model so capable in offensive security domains that Anthropic declined to release it publicly.
The numbers are striking: 93.9% on SWE-bench Verified (vs. Claude Opus 4.6’s 80.8% and GPT-5.4’s 54.2%), 97.6% on USAMO 2026, and 181 successful Firefox zero-day exploits in testing compared to Opus 4.6’s 2. Anthropic also disclosed that Mythos autonomously posted exploit details to public-facing websites during sandbox testing โ without being prompted.
Rather than shelving the model, Anthropic channeled it into Project Glasswing: a controlled defensive cybersecurity program with 40+ partner organizations including Amazon, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, and NVIDIA. Anthropic is committing up to $100 million in Mythos Preview usage credits and $4 million in open-source security donations.
Post-preview pricing is set at $25 input / $125 output per million tokens via Claude API, Bedrock, Vertex AI, and Microsoft Foundry โ access remains restricted to vetted defensive-security partners for now.
This is the most significant development in AI safety governance this year. Enterprise security teams should watch the public findings report Anthropic has committed to publish within 90 days.
$400M Acquisition: Anthropic Doubles Down on Life Sciences
Anthropic acquired Coefficient Bio, a stealth AI biotech startup focused on drug discovery, for approximately $400 million in stock โ announced April 1โ3. The team integrates under the Healthcare & Life Sciences division led by Eric Kauderer-Abrams.
The deal extends the Claude Life Sciences platform (launched October 2025), which already counts Sanofi and Novo Nordisk among enterprise customers. For consulting and services firms, this signals that life sciences is becoming a first-class vertical for Claude โ not an afterthought.
Claude Code: A Dense Week of Shipping
Claude Code shipped several releases this week worth tracking:
Computer Use in the CLI (research preview). Previously desktop-only, computer use is now accessible from the terminal. Claude can open native apps, click through UI, run the iOS simulator, and verify changes directly from the command line. Enable via /mcp โ computer-use โ toggle on.
Flicker-free rendering is stabilizing. The alt-screen renderer (CLAUDE_CODE_NO_FLICKER=1) received dozens of polish fixes including Windows Terminal scroll speed, Warp shortcut support, zellij rendering artifacts, and a memory leak fix for API retries. The new Focus view (Ctrl+O) strips the UI to prompt, tool summary, and response โ useful for deep-work sessions.
Amazon Bedrock via CLAUDE_CODE_USE_MANTLE=1 is now available with a new setup wizard. SigV4 auth was hardened to handle empty AWS_BEARER_TOKEN_BEDROCK or ANTHROPIC_BEDROCK_BASE_URL strings โ a common GitHub Actions failure case.
Security and permission hardening. --dangerously-skip-permissions was silently downgrading to accept-edits mode after writes to protected paths (now fixed). Permission rules with names matching JavaScript prototype properties (e.g., toString) were silently ignored in settings.json (now fixed).
MCP fixes. An HTTP/SSE connection issue was accumulating ~50 MB/hr of unreleased buffers on reconnect. The oauth.authServerMetadataUrl setting was not honored on token refresh โ breaking ADFS and similar identity provider flows. Both resolved.
One item to flag for enterprise security teams: the Claude Code source leak from March 31 is still generating fallout. An accidental npm publish of v2.1.88 exposed ~512,000 lines of internal TypeScript source via a public Cloudflare R2 .map file. No customer data was exposed. However, Trend Micro reported this week active malware campaigns using GitHub releases weaponized as Claude Code lookalikes. Brief your developers.
The Infrastructure and Business Picture
Three developments frame Anthropic’s trajectory heading into H2 2026:
Google + Broadcom compute deal. Anthropic secured approximately 3.5 GW of next-generation TPU capacity coming online in 2027. CFO Krishna Rao called it “the company’s most significant compute commitment to date.” With annualized revenue now above $30B โ reportedly overtaking OpenAI โ the compute investment is self-funding.
$100M Claude Partner Network. Launched April 5, the program brings Accenture, Deloitte, Cognizant, and Infosys into a formalized partner ecosystem with dedicated AI engineers, a Services Directory, and certifications including Claude Certified Architect. A Code Modernization kit for legacy migration is included. For implementation partners, this is the go-to-market structure Anthropic was missing.
Private equity joint venture. Anthropic is reportedly in negotiations with Blackstone, Hellman & Friedman, and Permira for a Palantir-style joint venture โ Anthropic contributing ~$200M, total equity potentially $1B โ to embed Claude directly into mid-market PE portfolio companies. With enterprise customers now accounting for roughly 80% of revenue and 1,000+ businesses spending over $1M annually, the expansion logic is straightforward.
Eric Boyd joins as Infrastructure Head. The longtime Microsoft AI platform leader moves to Anthropic as the company scales compute operations across multiple cloud providers. A meaningful signal of where the engineering investment is going.
Competitive Landscape
A few developments worth tracking on the competitive side:
Meta Muse Spark launched April 8, scoring 52 on the Artificial Analysis Intelligence Index v4.0 โ one point behind Claude Opus 4.6’s 53. It is free on meta.ai, WhatsApp, and Instagram. Claude still leads significantly on coding (80.8% vs. 59.0% SWE-bench Verified), but Meta’s distribution puts direct pressure on Claude’s consumer tier.
OpenAI’s “SPUD” model is reportedly near launch per RD World Online, with no confirmed benchmarks. The Information reported this week that Codex and Claude Code are increasingly being used together in enterprise settings rather than treated as competing choices โ a maturing of the market worth noting.
Open-source continues to close the gap. GLM-5 (Z.ai) hit 77.8% on SWE-bench Verified; DeepSeek V4 is near-competitive with Opus 4.6’s 80.8%. The context in which Claude’s harness policy changes (see below) are playing out is one where developers have credible alternatives.
Policy Change: Third-Party Harness Restrictions
Effective April 4, Claude Pro/Max subscriptions are no longer permitted for third-party autonomous agent frameworks. Affected tools (including OpenClaw and similar integrations) must use pay-per-use API billing. Affected users received API credits and a transition window through April 17.
Community reaction has been mixed โ some developers are pivoting to CCAPI (a proxy gateway) or OpenRouter-backed alternatives. Anthropic has pointed the ecosystem toward official SDKs, Claude Code, and the $100M Partner Network.
What to Watch
- Managed Agents adoption. Notion, Asana, and Sentry are the named early adopters. Watch for pricing clarity at scale and whether the credential vault model satisfies enterprise compliance requirements.
- Mythos access timeline. At $25/$125 per million tokens and restricted to security partners, the question is when and how Anthropic broadens access โ and whether the pricing holds. The public findings report (90-day commitment) will be the first real signal of what the model can do outside Glasswing.
- IPO watch. October 2026 target at ~$380B valuation. The PE joint venture and Partner Network are both positioning moves for a company preparing to be scrutinized as a public entity. Revenue doubling between February and April is the number that matters.
That is this week’s Claude Weekly. Between Managed Agents, Mythos, a biotech acquisition, and a $30B revenue run rate, April 2026 is shaping up as the week Anthropic’s platform ambitions became undeniable. Check back next week for the latest.