Anthropic AI Updates: June 3, 2026

1. Anthropic Expands Project Glasswing to 150 Partners Across 15+ Countries

Anthropic. Anthropic expanded Project Glasswing, its initiative to harden critical software, from roughly 50 initial partners to about 150 organizations across more than 15 countries. Partners span essential infrastructure sectors including power, water, healthcare, and communications, and use Claude Mythos Preview to surface vulnerabilities, reportedly identifying more than 10,000 high- or critical-severity flaws so far. The company frames the effort as preparing the cybersecurity industry ahead of broader availability of more capable AI models. Source

2. Claude Code Adds Dynamic Multi-Agent Workflows

Anthropic. Anthropic detailed how Claude Code now lets developers build custom multi-agent orchestration that adapts dynamically to a task by spawning specialized subagents with focused goals. The post describes reusable patterns such as classify-and-act, fan-out-and-synthesize, adversarial verification, and tournaments, aimed at migrations, deep research, verification, and large-scale triage. The approach is positioned to mitigate single-context failure modes like agentic laziness and self-preferential bias. Source

3. Anthropic Describes Running an AI-Native Engineering Org

Anthropic. Fiona Fung, director of engineering for Claude Code and Claude Cowork, described how Anthropic’s engineering process changed once agentic coding became the default. The team shifted from heavy upfront planning to just-in-time planning, moved routine style and bug review to AI while humans concentrate on domain expertise, and adjusted hiring toward creative builders and systems engineers over raw coding throughput. The piece argues that as code generation stops being the bottleneck, verification, security, and strategic human judgment become the new priorities. Source