Anthropic AI Updates: May 15, 2026

1. Anthropic forms $200 million partnership with the Gates Foundation

Anthropic. Anthropic announced a four-year, $200 million partnership with the Gates Foundation that combines grant funding, Claude usage credits, and technical support across global health, life sciences, K-12 education, agricultural productivity for smallholder farmers, and US economic mobility initiatives. The collaboration aims to accelerate vaccine and therapy development and build AI tools for classrooms in multiple regions, with Anthropic framing the work as extending AI benefits “in areas where markets alone will not.” Source

2. Anthropic publishes “2028” policy paper on US-China AI leadership

Anthropic. A new policy paper from Anthropic lays out two scenarios for global AI leadership by 2028: one where the US and allies hold a 12-24 month lead through tightened export controls, anti-smuggling enforcement, defense against model distillation attacks, and aggressive promotion of American AI globally; and one where China reaches near-parity through illicit chip access and large-scale model theft. The paper argues advanced semiconductors are the binding constraint on Chinese progress and recommends three priorities: closing chip-smuggling loopholes, defending against model theft, and championing democratic-aligned AI adoption abroad. Source

3. Anthropic ships “The Founder’s Playbook” for building AI-native startups

Anthropic. Anthropic released The Founder’s Playbook, a guide structured around four startup lifecycle stages (Idea, MVP, Launch, Scale) with frameworks, exercises, and case studies from founders at Ambral, Carta Healthcare, and HumanLayer. The resource is aimed at early-stage operators architecting their companies around Claude from day one rather than retrofitting AI later. Source

4. Anthropic publishes guide on running Claude Code in large codebases

Anthropic. A new Claude blog post argues that Claude Code navigates monorepos and legacy systems via agentic search rather than embedding-based RAG, so success depends on the surrounding harness (CLAUDE.md files, hooks, skills, plugins, MCP servers, and LSP integrations) more than the model itself. Recommendations include layering lean CLAUDE.md files per directory, excluding generated artifacts, reviewing project memory every 3-6 months as models improve, and assigning a dedicated owner (often in developer experience) to standardize Claude Code conventions before a broad rollout. Source

5. PwC expands Anthropic alliance, will train 30,000 professionals on Claude

Anthropic. PwC and Anthropic announced a major expansion of their alliance, rolling out Claude Code and Cowork starting with US teams and extending toward PwC’s global workforce of more than 364,000 across 136 countries, plus training and certifying 30,000 US professionals on Claude and standing up a joint Center of Excellence. The expansion targets three areas (building agentic AI tools for clients, deploying AI across dealmaking, and reinventing operating models) and introduces a Claude-native finance business group inside PwC’s Office of the CFO practice, with reported delivery improvements of up to 70% on early engagements. Source

6. Claude Code v2.1.142 adds background-session CLI flags and switches fast mode to Opus 4.7

Anthropic. Claude Code v2.1.142, tagged on 2026-05-14, introduces new claude agents command-line flags for configuring background sessions: --add-dir, --settings, --mcp-config, --plugin-dir, --permission-mode, --model, --effort, and --dangerously-skip-permissions. The release also changes fast mode’s default model from Opus 4.6 to Opus 4.7, with an environment variable override available for users who want to revert. Source