🧭 Claude Design Launches from Anthropic Labs — Conversational Visual Asset Generation for Non-Designers

Anthropic released Claude Design on April 17, an experimental product from the Anthropic Labs division that lets users generate professional visual assets — slide decks, one-pagers, wireframes, marketing collateral, and web prototypes — through natural-language prompts and inline conversational editing. It is distinct from Claude Artifacts (which generates standalone HTML/code snippets in a side pane) and is instead a purpose-built design-oriented product with its own export pipeline, adjustment controls, and team design-system ingestion. The research preview is available now to Pro, Max, Team, and Enterprise subscribers at no additional charge.

How it works

Claude Design is powered by Opus 4.7 and operates on a canvas-and-conversation model: you describe what you want, Claude generates a visual draft, and you refine it through follow-up prompts or on-canvas sliders (colour temperature, density, font weight, brand prominence). The product reads your company's existing codebase and design files to extract brand tokens — primary colours, typefaces, logo placement rules — and applies them automatically across generated assets. This means a generated slide deck for an enterprise customer will use that customer's brand kit without a manual style pass.

Supported output formats

• PDF — for print-ready deliverables and investor decks
• PPTX — editable PowerPoint for team iteration
• Standalone HTML — self-contained interactive presentations for web delivery
• Canva export — sends layers to a Canva project for pixel-level editing by designers

The Anthropic Labs label signals that the product is in active research preview with a real user base but without a committed GA timeline. Features will change based on feedback. The team is particularly interested in how enterprise customers use the design-system ingestion feature, since that is the most technically novel capability and the one most likely to require iteration before a broader rollout.

🧭 Claude for Financial Services Gets Claude for Excel, Seven Live Data Connectors and Six Pre-Built Agent Skills

Anthropic announced a significant expansion of its Claude for Financial Services vertical on April 17, adding three distinct new layers to the offering: a native Excel integration in beta, a set of live market-data connectors, and a library of pre-built Agent Skills targeting the most common research workflows in buy-side and sell-side finance. The expansion positions Anthropic directly against Bloomberg's Terminal AI layer and Microsoft Copilot for Finance for the professional financial services market.

Claude for Excel (beta)

Claude for Excel is a sidebar add-in that lets users describe what they want in natural language — "calculate the 3-year CAGR for each row in column C" or "flag any cell where revenue growth exceeds 40% year-over-year" — and Claude reads, analyzes, modifies, and creates Excel workbooks in response. It supports formula generation, conditional formatting, chart creation, and cross-sheet data summarisation. Available to Max, Enterprise, and Teams subscribers. The beta is limited to workbooks under 50MB and English-language cells; multi-language support is on the roadmap.

Seven new live data connectors

Claude for Financial Services can now pull live data from seven new sources, turning Claude into a single interface for research aggregation:

• Aiera — earnings call transcripts and event intelligence
• Third Bridge — expert network interview transcripts
• Chronograph — private equity portfolio monitoring
• Egnyte — internal document stores and deal room data
• LSEG (London Stock Exchange Group) — real-time market pricing and fundamentals
• Moody's — credit ratings and risk analytics
• MT Newswires — real-time financial news and regulatory filings

Six pre-built Agent Skills

Rather than asking financial professionals to build their own agents, Anthropic is shipping six production-ready Agent Skills that combine data connectors with orchestrated multi-step reasoning:

• Comparable company analysis — builds a comps table from LSEG + Moody's data for a given ticker
• DCF modelling — generates a five-year discounted cash flow model with stated assumptions
• Due diligence data packs — aggregates company intelligence from Aiera, Third Bridge, and Egnyte into a structured report
• Company profiles — one-page summaries combining public filings, earnings transcripts, and news
• Earnings analysis — extracts and compares guidance, beats/misses, and management tone across quarters
• Coverage report generation — drafts analyst-style initiation reports from connector data

🧭 Opus 4.7's Cyber Verification Program: How Security Researchers Get Allowlisted to Skip the New Safety Layer

Buried alongside the Opus 4.7 launch announcement is a detail that matters specifically to the security community: the new model ships with an automated real-time cybersecurity safety layer that intercepts and blocks high-risk requests — offensive exploit generation, novel malware development, and attack-infrastructure design among them. This is the first time Anthropic has deployed an always-on cybersecurity classifier at the model level rather than as an operator-configurable policy. And alongside it, Anthropic launched the Cyber Verification Program, an application-based allowlist that lets legitimate security professionals bypass those blocks.

What the safety layer blocks by default

Anthropic has not published an exhaustive list, but the announcement describes the classifier as targeting: novel vulnerability exploitation (requests that combine specific CVE information with working exploit generation), malware development for non-sandboxed use, attack-infrastructure design (C2 setup, phishing kit generation), and social-engineering content generation at scale. Standard security topics — CTF challenges, code auditing, explaining published CVEs, defensive hardening — are explicitly stated to remain unaffected.

Applying for the Cyber Verification Program

The application process requires:

• Proof of employment at a licensed penetration testing firm, a corporate red team, or a recognised vulnerability research organisation
• A description of the specific use cases requiring elevated access (e.g. "internal red team exercises under signed SoW with client X")
• Agreement to Anthropic's responsible-disclosure terms — researchers who discover model-level vulnerabilities through the program must disclose to Anthropic within 90 days

Approved applications result in an API-key-level flag that unlocks the full Opus 4.7 capability surface for requests from that key. There is no model-switch involved — the same model ID is used, with the classifier operating in a permissive mode for verified keys.

# Check whether your API key has Cyber Verification access
import anthropic

client = anthropic.Anthropic()

# Make a probe call — the response metadata will include
# a 'cyber_verified' flag in the model metadata block
# if your key is in the allowlist
response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=10,
    messages=[{"role": "user", "content": "ping"}]
)
print(response.model)           # claude-opus-4-7
# Check response headers for X-Anthropic-Cyber-Verified: true