Z.ai's open-source GLM-5.3 benchmarks at the frontier and has allegedly already exploited a real coding-tool vulnerability, raising dual-use alarm bells. ⚖️
Happy Saturday. Two OpenAI-adjacent stories dominate a quieter Friday news cycle: Anthropic's IPO valuation is apparently tied to an ambitious $190–200B revenue forecast for 2028, while Nvidia has quietly scaled back its headline-grabbing $250B data-center guarantee to OpenAI. Meanwhile, Z.ai's GLM-5.3 raises eyebrows by reportedly finding a serious vulnerability in coding tool Cursor, and the US is reportedly preparing to force allies to publicly choose sides in the AI race with China.
Two significant model-layer stories land today: Z.ai's GLM-5.3 pushes Chinese open-source models into sensitive cybersecurity territory, and Anthropic quietly discloses a next-generation model beyond Claude Mythos 5 while surfacing fresh alignment concerns.
Infrastructure investment signals mixed caution and commitment today: Nvidia has walked back its landmark $250B OpenAI data-center guarantee, while chipmaking equipment giant Lam Research is doubling down with $3B+ in new lab spending over five years.
Anthropic's IPO preparations dominate the funding conversation today, with Reuters reporting that its target valuation hinges on an eye-popping $190–200B revenue forecast for 2028 — a number that sets a very high bar for one of AI's most watched public market debuts.
Security and developer-tooling concerns take center stage at the middleware layer, with a new report flagging weak API controls as a leading enterprise threat in the agentic era, while a French startup pitches a fresh approach to GPU inference optimization.
OpenAI's enterprise momentum is affirmed at the C-suite level — its CFO telling investors that enterprise revenue now exceeds consumer — even as leadership turmoil casts a shadow over the company's IPO narrative.
US foreign policy is hardening around AI geopolitics — Reuters reports an imminent demand that allies choose sides in the AI race with China — while French publishers escalate their fight against Google's AI news summaries, and Indian courts are starting to grapple with AI-generated deepfake content.
Tech Visionary Says the Big AI Labs Don't Get What People Want — Tim O'Reilly built his empire by identifying technology shifts before they were obvious — his critique that frontier labs are optimizing for the wrong user outcomes is the kind of contrarian thesis a senior strategist should interrogate seriously. At a moment when IPO valuations, capex pledges, and model releases dominate the conversation, his open-source-first argument offers a useful corrective lens on where durable value may actually accrue. Read →