AI News Digest — 2026-04-28

Highlights

Musk vs. Altman head to court over OpenAI’s future: A Northern California trial opening this week could decide whether OpenAI may operate as a for-profit ahead of its IPO — and could even force out Altman.
OpenAI and Microsoft restructure their partnership: OpenAI is now free to sell its products on AWS and other clouds, ending Azure’s API exclusivity, while OpenAI-to-Microsoft revenue sharing continues.
GitHub Copilot shifts to token-based metered billing: With agentic usage frequently exceeding flat plan costs, Copilot is moving to consumption pricing — a likely end of the unlimited-use era for AI dev tools.
Fujitsu builds a “Physical OS” to unify physical AI: A new joint research center with Carnegie Mellon will feed into Fujitsu Kozuchi Physical OS, an integration layer for embodied AI systems.
Google tests AI chatbot search inside YouTube: A conversational search experiment now rolling out to some Premium users pulls together longform videos, Shorts, and explanatory text.

Robinhood account creation flaw abused to send phishing emails: Threat actors exploited Robinhood’s signup flow to inject phishing messages into legitimate platform emails, making fake “suspicious activity” alerts appear authentic.
GoDaddy mistakenly transferred a developer’s domain to a stranger: Web developer Austin Ginder reported that GoDaddy moved his domain to a third party — the registrar called the handover “a legitimate procedure” before recovery happened only via direct contact with the new holder.

Musk vs. Altman head to court over OpenAI’s future: With OpenAI’s IPO looming, the Northern California trial could rule on the legality of its for-profit conversion and may even force leadership changes.
OpenAI–Microsoft partnership restructured: OpenAI gains the right to distribute via AWS and other clouds, ending Azure-exclusive API delivery; Microsoft’s revenue share to OpenAI ends, though OpenAI’s share to Microsoft continues, accelerating a multi-cloud strategy.
Google tests AI chatbot search for YouTube: A new “AI Mode-like” experimental search experience surfaces longform videos, Shorts, and text answers in a conversational format.
GitHub Copilot moves to token-metered billing: As agentic workflows routinely exceed flat-plan costs in just a few requests, Microsoft is repricing Copilot on consumption to close the gap between price and actual compute cost.

Fujitsu develops a “Physical OS” to bundle physical AI: Fujitsu and Carnegie Mellon University have set up a joint physical-AI research center; results will feed Fujitsu Kozuchi Physical OS, a new platform aimed at unifying physical AI components.
Fourth-year “Small-Start AI Pavilion” shifts toward MPU-class silicon: At the 10th AI/AI EXPO Spring, ST Microelectronics, NXP, Nuvoton, and Renesas showcased low-power AI inference on microcontrollers and MPUs — with the lineup tilting more heavily toward MPUs than in prior years.
“Niijima” fails asking AI for research — generative-AI cautionary manga: An ITmedia comic walks through the “land mines” a new hire trips while introducing generative AI to a corporate workflow, framing first steps for enterprise AI use.

OpenAI’s commercial structure under pressure on two fronts: a courtroom challenge to its for-profit conversion (Musk lawsuit) and a renegotiated Microsoft deal that opens the door to AWS and beyond.
Pricing AI tools at their real cost: GitHub Copilot’s pivot to token-metered billing signals that flat-rate “all-you-can-eat” plans don’t survive contact with agentic workloads.
Physical / embodied AI as a platform play: Fujitsu’s Physical OS push and the MPU-heavy edge-AI showcase point to silicon-and-runtime stacks becoming the new battleground beyond pure model APIs.
Trust and identity infrastructure are still soft targets: Robinhood’s signup-flow phishing abuse and GoDaddy’s mistaken domain transfer both show that high-trust account and identity systems remain exploitable through process gaps, not just code bugs.

For detailed summaries of selected research papers, see papers.md.