✅ Responsible Scaling Policy v3.0 — Full Text Published

Anthropic has published the full text of Responsible Scaling Policy v3.0, following the policymaker pre-briefings conducted earlier this week. The document is the most substantial revision of the RSP since the ASL framework was introduced in 2025, and introduces three significant structural changes to how Anthropic governs its model deployment process.

The three structural changes in RSP v3.0

• Deployment Review Board (DRB) — a new internal governance body with authority to delay any model release pending additional safety evaluation. The DRB comprises representatives from research, policy, legal, and external safety advisors, and must reach consensus before any ASL-3-or-above candidate proceeds to deployment review
• Mandatory third-party audit for ASL-3 candidates — models that trigger ASL-3 screening criteria must undergo evaluation by at least two independent external auditors before the DRB review. Previously, third-party audit was voluntary and aspirational
• ARARA evaluation as a hard gate — the new autonomous replication and resource acquisition evaluation cluster introduced in the February Risk Report is now a mandatory hard gate at every model release, not an advisory check

Anthropic has also updated the language around ASL-3 trigger criteria to adopt the revised uplift thresholds from the February Risk Report. The full document is available at anthropic.com/research/rsp-v3.

✅ Enterprise Agents Briefing — Cowork Integration Patterns for Large Organisations

Anthropic held its first Enterprise Agents Briefing today — a closed session for enterprise customers covering best practices and roadmap previews for deploying Cowork at organisational scale. The session was attended by IT and AI leads from approximately 40 Enterprise customers and is being followed up with written guidance.

The key integration patterns covered included: configuring Cowork's workspace folder designation across Windows and macOS device fleets via MDM; using the Enterprise admin console to set default MCP connector permissions for all users in a workspace; and designing CLAUDE.md templates that enforce consistent agent behaviour across a department or team. Anthropic also previewed a forthcoming Cowork Policies API — a programmatic interface for administrators to set and enforce what Cowork is and is not permitted to do with files, external services, and user data — expected to reach preview in Q2.