Elon Musk returned to the stand in Oakland federal court for a third day, enduring aggressive cross-examination from OpenAI lead attorney Bill Savitt, who argued Musk's lawsuit is motivated by competitive spite, not principle. Musk testified that OpenAI's leaders wanted to "get rich" off a nonprofit mission, and is seeking $130 billion in damages plus reversion of OpenAI to nonprofit status and removal of Altman and Brockman from the board. The jury's advisory verdict will guide Judge Yvonne Gonzalez Rogers on a decision that could fundamentally reshape the most important AI company in the world.
Microsoft and OpenAI formally restructured their seven-year exclusive partnership, making OpenAI's models — including the just-shipped GPT-5.5 — available on Amazon Bedrock for the first time. Microsoft's license to OpenAI IP continues through 2032 but shifts from exclusive to non-exclusive, while OpenAI gains freedom to distribute across any cloud provider. This is the single biggest structural shift in AI cloud distribution since the original 2019 deal.
A paper presented at ICLR 2026 in Rio de Janeiro demonstrates that training models for stronger reasoning via reinforcement learning increases tool-hallucination rates proportionally to task gains. OpenAI's own data shows o3 hallucinating on 33% of queries — more than double o1's 16% rate. The authors frame this as a "fundamental reliability-capability trade-off," meaning current reasoning-enhancement methods were never designed to jointly optimize accuracy and tool restraint. This has severe implications for agentic AI deployment.
The White House is developing executive guidance that would let federal agencies sidestep Anthropic's supply-chain risk designation — imposed after Anthropic refused to ease restrictions on domestic surveillance and autonomous weapons use — and onboard Mythos, the company's frontier model with reportedly unprecedented cybersecurity capabilities. Trump signaled a thaw, telling CNBC that Anthropic is "shaping up" and can "be of great use."
Alphabet, Microsoft, Amazon, and Meta all report after the bell today within an 80-second window, with combined 2026 AI infrastructure commitments approaching $600 billion (Amazon ~$200B, Alphabet ~$180B, Meta ~$125B, Microsoft on a $120B annualized pace). Investors are laser-focused on whether this unprecedented spending is generating proportional revenue, especially after reports that OpenAI missed its own user acquisition targets.
Sergey Brin, now writing code daily in full "founder mode," has assembled a dedicated DeepMind strike team led by Gemini pre-training veteran Sebastian Borgeaud with a single mandate: close the widening coding capability gap with Anthropic's Claude. The immediate trigger was Anthropic's Claude Opus 4.7 release on April 16. Brin's internal memo calls enhanced coding ability the "key step" toward AI self-improvement. The team will train on Google's 2-billion-line internal codebase.
Google committed $10 billion upfront with up to $30 billion more contingent on performance milestones, valuing Anthropic at $380 billion. The deal provides Anthropic with infrastructure to address capacity constraints from surging enterprise and consumer demand for Claude. Multiple analysts called it a "screaming bargain" given Anthropic's growth trajectory, while others noted it effectively ends the AI "Big Three" era by isolating OpenAI.
OpenAI officially shut down Sora's web and app on April 26 (API follows in September), ending a product that peaked at ~1 million users but burned roughly $1 million per day in compute. The move signals OpenAI's strategic pivot away from creative consumer tools and toward coding, enterprise products, and a planned "super app" consolidating ChatGPT and other services. Users have a final export window before all data is permanently deleted.
Anthropic's annualized revenue reached $30 billion in April — up from $9 billion four months ago — making it the fastest-growing company in American business history by this measure. Over 1,000 enterprise customers now spend $1M+ annually, and Claude Code business subscriptions have quadrupled since January. OpenAI's CRO disputed the figure, arguing ~$8B is overstated due to gross vs. net accounting of cloud partner revenue.
Nvidia, Alibaba, and Tencent each released world models in the past two weeks, joining a wave of investment in AI systems trained on physical-world data that can model 3D environments, physics, and spatial reasoning — capabilities today's chatbots lack entirely. AMI Labs raised over $1B (a European record first round) for a radical world-model approach. Nature notes these systems still fail at basic physics predictions, raising safety questions for robotics and autonomous vehicles.
