Anthropic AI Updates: April 16, 2026

1. Bloomberg Reveals How Anthropic Discovered Mythos Was Too Dangerous for Public Release

Anthropic. A Bloomberg feature details how Anthropic’s internal researchers discovered that Claude Mythos could hack critical computing systems, finding a 27-year-old bug in OpenBSD and a 16-year-old flaw in FFmpeg that survived five million automated security tests. Rather than releasing publicly, Anthropic launched Project Glasswing, a defensive consortium with Amazon, Apple, Microsoft, and CrowdStrike, committing $100 million in credits. Source

2. UK AI Security Institute Publishes Mythos Cyber Capabilities Evaluation

Anthropic. The UK AI Security Institute published its evaluation of Claude Mythos Preview, reporting a 73% success rate on expert-level capture-the-flag tasks and full completion of its 32-step enterprise network attack range on 3 of 10 attempts. The range simulates a complete network takeover that typically takes human experts around 20 hours, and the Bank of England is convening a briefing in response. Source

3. Canada’s AI Minister Praises Anthropic’s Responsible Mythos Approach

Anthropic. Canada’s AI minister Evan Solomon praised Anthropic’s approach to Mythos, stating “Working with defenders first, rather than releasing this new model broadly, is the responsible path.” Canada’s finance minister also plans to discuss Mythos with US Treasury Secretary Scott Bessent at IMF/World Bank meetings in Washington. Source

4. TrendAI Partners with Anthropic for AI Security Operations

Anthropic. TrendAI (Trend Micro’s AI brand) announced a strategic partnership with Anthropic, embedding Claude models across its platform for agentic workflows, automation, AI-native security operations, and threat research. The partnership spans the full AI security lifecycle, from vulnerability discovery to operational response. Source

5. Claude Code v2.1.109-110: TUI Mode, Push Notifications, and Focus Command

Anthropic. Two Claude Code releases shipped on April 15. v2.1.110 added the /tui command for flicker-free fullscreen rendering, push notification support via Remote Control, a new /focus command, autoScrollEnabled config, and improved plugin management. v2.1.109 introduced a rotating progress hint for the extended-thinking indicator. Source

6. Claude Experiences Multi-Hour Outage Across All Services

Anthropic. Claude.ai, the API, and Claude Code experienced a significant outage on April 15, with over 7,000 users reporting problems. Anthropic confirmed elevated errors at 10:53 AM ET, issued a partial fix, saw recurrence at 11:40 AM, and fully resolved the incident by 1:42 PM ET. The disruption was attributed to surging demand following recent major model releases. Source

7. Anthropic Opens 2026 Fellows Program Applications

Anthropic. Anthropic opened applications for May and July 2026 cohorts of its Fellows Program, providing a $3,850/week stipend, approximately $15k/month in compute credits, and direct mentorship for researchers working on safety and security projects for four months. Over 40% of previous fellows subsequently joined Anthropic full-time, with applications due April 26. Source