AI News Roundup: Nvidia’s $40B AI Bets, Anthropic’s Mythos Shakes Cybersecurity, GPT-5.5 Goes Default

Nvidia Tops $40 Billion in AI Equity Bets, Anchored by $30B OpenAI Stake

Nvidia has already committed more than $40 billion in equity investments in 2026, cementing its transformation from chipmaker to full-stack AI kingmaker. The headline deal: a $30 billion stake in OpenAI, giving Nvidia significant influence over the company building some of the most advanced AI systems in the world.

This week alone, Nvidia agreed to invest up to $3.2 billion in glass maker Corning and $2.1 billion in data center operator IREN. The remaining $10 billion-plus is spread across seven multi-billion-dollar deals in publicly traded companies and roughly two dozen private startup rounds — all targeting companies up and down the AI infrastructure stack.

Critics argue the strategy creates a circular economy: Nvidia invests in companies that then spend that capital buying Nvidia chips. CFO Colette Kress countered on the most recent earnings call that the company invests where it sees a need to ensure compute capacity is being built around its hardware. Either way, the scale is unprecedented for a semiconductor company.

Anthropic’s Mythos Sparks Cybersecurity “Hysteria” After Finding Tens of Thousands of Zero-Days

Anthropic’s restricted-access Mythos model has identified tens of thousands of zero-day vulnerabilities across every major operating system and web browser, with over 99% still unpatched. For perspective, an earlier Anthropic model found roughly 20 vulnerabilities in Firefox — Mythos found nearly 300 in that browser alone.

CEO Dario Amodei warned of a “moment of danger” as the model exposes the fragility of software infrastructure that underpins global finance and critical systems. Anthropic has restricted access to a small group of launch partners including Apple, Amazon, JPMorgan Chase, and Palo Alto Networks through its Project Glasswing initiative, with access extended to over 40 additional organizations that maintain critical infrastructure.

Some cybersecurity experts remain skeptical, telling CNBC that similar vulnerabilities can be found using existing models with clever orchestration. But the sheer scale of Mythos’s output — and the speed at which it produces results — has nonetheless triggered an industry-wide scramble to patch before exploit code emerges.

OpenAI Releases GPT-5.5 Instant as the New Default ChatGPT Model

OpenAI rolled out GPT-5.5 Instant as the default model for all ChatGPT users, replacing GPT-5.3 Instant. The new model reduces hallucinated claims by 52.5% on high-stakes prompts covering medicine, law, and finance — a meaningful reliability improvement for the hundreds of millions who use ChatGPT daily.

Benchmark gains are substantial: GPT-5.5 Instant scored 81.2 on the AIME 2025 math test (up from 65.4) and 76 on MMMU-Pro multimodal reasoning (up from 69.2). Enhanced personalization drawing from past chats, files, and connected Gmail is rolling out to Plus and Pro users, with broader availability coming soon.

Google I/O 2026 Preview: Gemini 4, 42.5 Exaflop TPUs, and AI Glasses

Google I/O arrives May 19 with what may be the company’s most ambitious AI lineup yet. The centerpiece is Gemini 4, reportedly scoring 84.6% on the ARC-AGI2 benchmark with a 2-million-token context window — large enough to fit entire codebases without retrieval augmentation.

On the hardware side, Google will unveil Ironwood TPUs delivering 42.5 exaflops of compute. The event will also showcase AI glasses built in partnership with Warby Parker, a robotics deal putting Gemini inside Boston Dynamics’ Atlas, and “Personal Intelligence” — a system connecting Gemini to Gmail, Photos, Drive, Calendar, and YouTube history, rolling out to 2 billion users across 200+ countries.

Novo Nordisk Partners with OpenAI to Bring AI Across Its Entire Business

Danish pharma giant Novo Nordisk announced a strategic partnership with OpenAI to integrate AI from drug discovery to supply chain operations. The collaboration will use advanced AI to analyze complex datasets, identify promising drug candidates, and compress the timeline from research to patient use — critical as Novo fights to regain market ground against Eli Lilly in the obesity and diabetes treatment space.

Pilot programs are already underway across R&D, commercial, and manufacturing operations, with full integration expected by end of 2026. OpenAI will also upskill Novo’s global workforce on AI literacy. The partnership includes strict data protection and human oversight governance frameworks.

US Government Secures Pre-Release Access to Test AI Models

The Center for AI Standards and Innovation struck agreements with Google DeepMind, Microsoft, and xAI allowing the U.S. government to evaluate AI models before public release. The NSA and Office of the National Cyber Director will lead safety testing for frontier models, developing pre-release oversight procedures without a direct veto over launches.

The framework represents the most concrete AI oversight mechanism the U.S. has implemented to date, though notably Anthropic and Meta are absent from the initial agreements. The move comes as AI-enabled threats accelerate — Mandiant’s M-Trends 2026 report found that exploits now routinely arrive before patches, with 28.3% of CVEs exploited within 24 hours of disclosure.