AI DAILY BRIEFING — Episode: General AI News Date: May 23, 2026 Duration: ~3-4 minutes spoken --- Good morning, and welcome to the AI Daily Briefing for Saturday, May 23rd, 2026. I'm Bob, and here's what's happening in AI. --- Anthropic's Project Glasswing has dropped a bombshell. In its first month of operation, the company's Claude Mythos Preview model — a specialized, unreleased AI tuned for cybersecurity — has uncovered over 10,000 high- or critical-severity vulnerabilities across essential software. This is not a lab exercise. Over 50 organizations, including Cloudflare, Mozilla, Oracle, and Microsoft, participated in the coordinated disclosure program. Cloudflare alone found roughly 2,000 bugs, 400 of them high or critical, with a false-positive rate better than its top human testers. Firefox 150 incorporated 271 Mythos-discovered fixes. The scale here is staggering — vulnerability discovery is no longer the bottleneck. Triage, patching, and disclosure are. And Anthropic is keeping Mythos-class models under wraps for now, because the same capabilities that find bugs could also exploit them. In other Anthropic news, the company is reportedly in early talks with Microsoft to rent Azure servers running Microsoft's own Maia 200 AI chips. This comes even as Anthropic has that massive $15 billion per year SpaceX capacity deal on the table. The Maia chips are designed for inference — running models rather than training them — which fits Claude's growing serving needs. The relationship between Microsoft and Anthropic remains one of the stranger dynamics in AI: partners, competitors, and occasional frenemies. Nvidia's numbers are in, and they are mind-bending. Q1 fiscal 2027 revenue hit a record $81.6 billion, up 85% year-over-year. Data center revenue alone was $75.2 billion — that is 92% of the total and up 92% from last year. The company guided $91 billion for Q2. They also announced an $80 billion share buyback authorization and raised the dividend. The AI infrastructure buildout is not slowing down. In policy news, President Trump delayed signing an AI executive order at the last minute this week, saying he quote "didn't like certain aspects of it" and didn't want anything that could be a "blocker" for AI jobs and American leadership over China. Separately, the Take It Down Act is now fully in effect, requiring platforms to remove AI-generated deepfakes and non-consensual intimate imagery within 48 hours. Two arrests were already made in Brooklyn under the act's criminal provisions this week. Microsoft announced plans to train one million AI professionals in Japan by 2030, partnering with NEC, NTT Data, SoftBank, Hitachi, and Fujitsu. It's also investing roughly 10.5 billion dollars in Japanese data centers and infrastructure. On the cultural front, Spotify is rolling out AI-powered podcast Q&A and personalized briefings, and will soon let authors auto-generate AI-narrated audiobooks. CapCut editing tools are coming to Google Gemini. And OpenAI launched a ChatGPT integration for Microsoft PowerPoint — generate or edit presentations from natural language prompts. Available now in beta for most ChatGPT plans. Grok AI usage stats: 117 million monthly active users as of March 31st, out of 550 million total monthly users across Grok and X. That's 21% adoption — a number that should make every other AI lab pay attention. And one note worth watching: anti-AI sentiment is rising in the US, driven by data center energy costs, water consumption in Texas, job displacement fears, and concerns about children's mental health. The backlash is real and it's growing. That's the briefing for May 23rd. Check the show notes for links to all the stories. I'm Bob — back tomorrow with more. --- Sources: X posts from @AnthropicAI, @testingcatalog, @nikkeibpITpro, @muskonomy, @penpenguin2023; The Verge; Ars Technica; Politico; Wall Street Journal.