Daily News · 2 min read

Anthropic AI Updates: May 13, 2026

anthropic claude claude-code managed-agents legal security

1. Code w/ Claude SF 2026 ships Dreaming, multiagent orchestration, Outcomes, and webhooks for Managed Agents

Anthropic. At its Code w/ Claude SF 2026 conference, Anthropic announced four new Claude Managed Agents capabilities: Dreaming, a scheduled memory-refinement process that mines past sessions for patterns between runs; multiagent orchestration, which lets a lead agent delegate to parallel specialist subagents with independent models, prompts, and tools; Outcomes, a developer-defined rubric system where a grader evaluates results and prompts revision (up to a 10-point gain on hard tasks); and webhooks that fire when defined outcomes complete. The company also doubled Claude Code rate limits and raised Claude Opus API limits, with follow-on events scheduled for London and Tokyo. Source

Anthropic. Anthropic introduced a legal-industry offering built on Claude Opus 4.7, shipping 20+ MCP connectors to platforms including Ironclad, Docusign, Definely, iManage, NetDocuments, Relativity, Everlaw, Consilio, Thomson Reuters CoCounsel, and Harvey, alongside 12 practice-area plugins spanning commercial, corporate, employment, privacy, regulatory, AI governance, IP, and litigation work. Claude now operates inside Word, Outlook, Excel, and PowerPoint with context preserved across apps, and an Access to Justice initiative partners with Free Law Project and legal aid groups via the Claude for Nonprofits program. Source

3. Anthropic’s security team built CLUE, a Claude Code threat detection platform that cut alert false positives from 33% to 7%

Anthropic. Anthropic detailed CLUE (Claude Looks Up Evidence), a natural-language threat detection platform its security team built with Claude Code: a proof of concept in a day, design and implementation within a week, and false positive rates on alerts dropped from roughly 33% to 7% after CLUE Triage shipped. Over a 30-day window CLUE handled about 12,000 queries and 27,000 tool calls (an estimated 1,870 hours of manual work avoided), turning multi-hour investigations into 3–4 minute queries by tying tool use into Slack, internal docs, and the data warehouse. Source