🧭 Anthropic Merges Claude Chat and Cowork Into One Interface — and Launches Claude Docs and Claude Slides

Anthropic announced a sweeping product consolidation rolling out to Pro and Max plan users from September 17, with Team and Free to follow in the coming weeks. The separate Claude Chat and Claude Cowork surfaces are being unified into a single conversation window: users will no longer choose between "chat mode" and "Cowork mode" at session start. Two new beta tools arrive alongside the merge — Claude Docs (collaborative document editing embedded inside conversations) and Claude Slides (AI-generated presentations exportable as PowerPoint or PDF). Claude Design also moves into the unified conversation flow.

What the unified interface changes for power users

Previously, starting a Cowork session required a separate entry point, which meant that a casual chat question couldn't seamlessly escalate into a collaborative artefact without reopening in a different mode. The unified interface resolves this: any conversation can spawn a Doc, a Slide deck, or a Design canvas at any point. Background task continuity is also new — Claude can continue processing after the user closes their laptop, with results resumable from any device. Recurring tasks can be scheduled to run automatically within conversations, moving Claude closer to a persistent personal assistant rather than a reactive chatbot.

Claude Docs and Claude Slides in beta

• Claude Docs: A shared, editable document that evolves inside the conversation. Multiple team members can be invited to a document thread; Claude can rewrite, restructure, or extend sections on request while preserving the collaborative edit history.
• Claude Slides: Generates a presentation from a prompt or an existing Doc. Output can be exported as .pptx or PDF. Slide design inherits Claude Design's brand-guardrail system when running in an Enterprise account.
• Persistent background tasks: Async task completion across device sleep. Results are pushed as notifications and accessible from any session, including mobile.

🧭 Novo Nordisk and Anthropic Partner to Accelerate Drug Discovery with Claude Science

Novo Nordisk — the Danish pharmaceutical company behind Ozempic and Wegovy — announced on September 16 a formal partnership with Anthropic to integrate Claude into its drug discovery and R&D workflows. The collaboration centres on Claude Science, Anthropic's AI workbench for scientists (launched in beta in June 2026), which provides built-in connectors to biological databases, computational chemistry tools, and large-scale compute environments. Novo Nordisk will deploy Claude Science across biological reasoning tasks and AI-accelerated software development internally, making this one of the largest pharmaceutical commitments to Claude as a scientific collaborator announced to date.

What Claude Science provides to Novo Nordisk researchers

• Database-connected reasoning: Claude Science includes pre-wired connectors to protein databases, molecular property databases, and clinical trial registries, letting researchers ask natural-language questions that span structured scientific data without writing database queries.
• Compute integration: The workbench can dispatch jobs to HPC clusters and cloud GPU pools directly from the conversation interface, compressing the iteration loop between hypothesis and simulation result.
• Collaborative research threads: Teams of scientists can work on shared Claude Science sessions, branching hypotheses and comparing reasoning chains across different biological targets.
• AI-accelerated development: Beyond wet-lab research, Novo will use Anthropic's frontier models to speed internal software development — from bioinformatics pipelines to clinical data management tools.

🧭 Inference Hooks: Anthropic's Inline Data Loss Prevention Layer for Claude Enterprise

Anthropic has moved Inference Hooks into wider beta for Claude Enterprise accounts. The feature routes every governed prompt through the customer's own external AI security server before Claude processes it — the security server can inspect, allow, or deny any prompt or tool-call response in real time. A denied request never reaches the model, and the user sees exactly which policy blocked it. Crucially, a single org-level setting covers the entire Claude surface: claude.ai, Claude Code, Claude Cowork, Docs, MCP connectors, Skills, and Plugins — eliminating the patchwork of per-surface DLP integrations that enterprise security teams have had to maintain.

How Inference Hooks work architecturally

When a user submits a prompt in any Claude surface, Claude's request path makes an outbound call to the customer-configured inspection endpoint before forwarding to the model. The inspection server receives the full prompt text (and any tool-call parameters), evaluates it against the organisation's policies, and returns an allow or deny signal with an optional user-facing policy message. The entire hook round-trip is synchronous and must complete within a configurable timeout before the model call proceeds or is blocked.

# Minimal Python inspection server (FastAPI)
from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()

class HookPayload(BaseModel):
    prompt: str
    session_id: str
    surface: str  # "claude.ai" | "claude-code" | "cowork" | "mcp" | ...

@app.post("/inspect")
def inspect(payload: HookPayload):
    # Example: block SSN patterns
    import re
    if re.search(r'\b\d{3}-\d{2}-\d{4}\b', payload.prompt):
        return {"action": "deny", "message": "Potential PII detected — SSN pattern blocked by org policy."}
    return {"action": "allow"}

Supported DLP vendor integrations

• Zscaler — routes Inference Hooks traffic through Zscaler Internet Access for unified policy enforcement
• Palo Alto Networks — integrates with NGFW and Prisma Access policy engines
• Netskope — connects to Netskope's AI transaction visibility and CASB layer
• Proofpoint — applies email DLP policy libraries to Claude prompt traffic