🧭 Anthropic Agrees to Brief the Financial Stability Board on Claude Mythos Cyber Risks

Anthropic has agreed to brief the Financial Stability Board (FSB) — the G20's global financial risk watchdog, chaired by Bank of England Governor Andrew Bailey — on cyber vulnerabilities uncovered by its restricted Claude Mythos Preview model. The briefing was requested by Bailey directly, reflecting growing concern among non-US regulators about systemic risks from a model that has been quietly surfacing decades-old zero-day vulnerabilities at scale.

What Mythos has found

• Thousands of high-severity vulnerabilities across major operating systems and browsers — including nearly 300 in Firefox alone, compared to roughly 20 found by prior Claude models running the same task.
• Restricted rollout — Mythos is distributed to approximately 40 organisations, mostly US-based: Amazon, Microsoft, CrowdStrike, JPMorgan Chase and others. The White House asked Anthropic to limit access pending broader policy review.
• Not a cyberweapon — Mythos is a vulnerability-discovery model, not an exploit generator. All findings are reported to vendors before any disclosure, following coordinated-disclosure norms. But the FSB's concern is that Mythos intelligence, if leaked or misused, could provide attackers a map of unpatched critical systems.

Why the FSB is watching

Bailey's concern — quoted in the Financial Times — centres on whether "this new version of the product is going to be able to, in a sense, identify vulnerabilities in other systems which can be exploited for cyber attack purposes." Financial infrastructure (payment rails, clearing houses, core banking systems) runs on layers of legacy software that Mythos-class models may expose faster than the sector can patch.

🧭 Claude for Legal Launches: 20+ MCP Connectors and 12 Practice-Area Plugins

Anthropic's largest vertical push to date arrived on May 12: Claude for Legal, a suite of more than 20 MCP connectors linking Claude directly to the software legal teams actually use, plus 12 practice-area plugins covering specific workflow types from M&A diligence to AI governance. The launch makes Claude the first major frontier model to ship a dedicated legal integration layer built on MCP as the standard.

The connector roster

The MCP connectors span the full legal software stack:

• Document platforms: iManage, NetDocuments, Box, LSuite
• Contract lifecycle: DocuSign, Ironclad
• Legal research: LexisNexis, Thomson Reuters (CoCounsel), Free Law Project (CourtListener)
• Litigation support: Everlaw

Each connector lets Claude query live data and take actions — drafting, reviewing, cross-referencing — without the user copying and pasting documents into the chat window.

The 12 practice-area plugins

Plugins add pre-configured skills, system prompts, and workflow checkpoints tuned to specific practice types: Commercial Legal, Corporate Legal (M&A diligence, closing checklists), Employment, Privacy, Product, Regulatory, AI Governance, IP, and Litigation, among others.

Who's deploying

• Freshfields is rolling Claude out across thousands of users and co-developing AI-native legal workflows with Anthropic.
• Thomson Reuters expanded its partnership to connect CoCounsel Legal — its flagship legal AI — to Claude via MCP, allowing CoCounsel users to invoke Claude reasoning inside their existing research interface.

🧭 Claude Code Agent View: One Dashboard for All Your Parallel Sessions

Released as a Research Preview with Claude Code v2.1.139, Agent View is a unified CLI dashboard that replaces the chaos of multiple terminal tabs when you're running parallel Claude Code sessions. Press ← (left arrow) from any session to enter a single list showing every active session — its state, last response, timestamp, and whether it's waiting for your input.

The core workflow

• Session table — each row is one Claude Code session. Columns: session name, current status (running / waiting / done), latest output snippet, time since last activity.
• Peek panel (Space) — select a row and press Space to see what the session needs, its recent output, and any pull requests it opened — without attaching to the full transcript. Most approvals can be handled from here.
• Dispatch new agents — launch fresh sessions directly from Agent View without opening a new terminal.
• Keyboard-first navigation — designed to stay in the terminal; no mouse required.

Who it's for

Agent View is available on Pro, Max, Team, Enterprise, and API plans running Claude Code v2.1.139 or later. It's most useful when you're running three or more concurrent agentic tasks — feature branches, test runs, documentation updates — that each need occasional human review but mostly proceed on their own.

# Ensure you're on v2.1.139+
claude --version

# From any session, press ← to open Agent View
# Or launch directly:
claude agent-view