AI News Weekly Summary — March 23–29, 2026 | Court Blocks Anthropic Ban

Court blocks Pentagon’s Anthropic ban. OpenAI shuts down Sora, loses Disney. Trump forms AI council. Huawei 950PR chip. Full AI roundup for March 23–29, 2026.

---

1. Federal Court Blocks Pentagon’s Anthropic Ban — “Orwellian” Ruling

On March 26, US District Judge Rita Lin in California issued a preliminary injunction blocking Anthropic’s classification as a “supply chain risk.” In a stinging 43-page ruling, she wrote:

“Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government.”

The government has 7 days (until ~April 2) to seek an emergency stay from the Ninth Circuit. CNBC: the court explicitly cited “First Amendment retaliation.”

This is a landmark ruling. For the first time, a federal court has found that the US government cannot punish an AI company for refusing to remove ethical safeguards. A precedent for the entire industry.

2. Altman Claimed He Tried to “Save” Anthropic — Leaked Slack Messages

On March 26, Axios published internal Sam Altman Slack messages from OpenAI (dated February 24 – March 2). Altman told employees he was “trying to save” Anthropic in the Pentagon dispute — while simultaneously seeking an opportunity to sign his own deal with Pentagon CTO Emil Michael.

On February 26, Altman sent an all-staff message claiming OpenAI “shares Anthropic’s red lines” and wanted to de-escalate. Privately, he wrote that Dario Amodei had been trying to undermine him for years.

Amodei had previously called OpenAI’s Pentagon messaging “straight up lies” and “safety theater.”

The Slack leak shows Altman torn between peacemaker and beneficiary. OpenAI walked out of Anthropic’s crisis with the Pentagon contract — and now must face that publicly.

3. OpenAI Shuts Down Sora — $15M/Day Costs, $2.1M Total Revenue

On March 24, OpenAI announced it’s discontinuing Sora — its AI video generator. The exact shutdown timeline has not been officially confirmed (unverified reports suggest the app on April 26, the API on September 24). No new API keys are being issued.

The numbers: estimated inference cost of $15 million per day, total Sora revenue — just $2.1 million. The Sora team transitions to world simulation research and robotics.

4. Disney Pulls $1B Investment in OpenAI After Sora Shutdown

Sora’s shutdown kills the Disney partnership. A three-year licensing deal — access to 200+ Disney, Marvel, Pixar, and Star Wars characters for Sora — collapsed in 90 days before any money changed hands.

Disney had planned to invest $1 billion in OpenAI and launch a curated Sora video channel on Disney+. Disney spokesperson: “We respect OpenAI’s decision to exit the video generation business.”

A billion-dollar deal collapsed in a quarter — with no funds transferred. This shows how fast AI partnerships can unravel when the economics don’t work.

5. Trump Appoints Tech Advisory Council — Zuckerberg, Huang, Ellison; No Musk or Altman

On March 25, Trump appointed 13 members of the President’s Council of Advisors on Science and Technology (PCAST): Mark Zuckerberg (Meta), Jensen Huang (NVIDIA), Larry Ellison (Oracle), Marc Andreessen, Sergey Brin (Google), Michael Dell, Lisa Su (AMD), Safra Catz (Oracle), Patrick Collison (Stripe), Dario Amodei (Anthropic), Alexandr Wang (Scale AI), John Jumper (Google DeepMind), and Yann LeCun (Meta).

Co-chairs: David Sacks (Trump’s AI and crypto advisor) and Michael Kratsios (science and technology policy director). The council will advise on AI, chips, workforce strategy, and national security.

Notably absent: Elon Musk and Sam Altman — their exclusion is widely discussed.

6. Apple Will Open Siri to Third-Party AI Services in iOS 27

On March 26, Bloomberg reported: Apple plans to open Siri to external AI assistants — Gemini, Claude, and others — as part of iOS 27. An “Extensions” feature will let apps integrate their agents with Siri.

Until now, ChatGPT had exclusive Siri integration. iOS 27 ends that monopoly. Apple positions the iPhone as an AI platform — independent of any single model provider.

7. Huawei 950PR — ByteDance and Alibaba Place Orders

March 27, Reuters: ByteDance and Alibaba plan to order the Huawei Ascend 950PR — a new AI chip designed as an NVIDIA alternative for the Chinese market.

The chip is more CUDA-compatible than the previous Ascend 910C and offers better response times. Price: 50,000 yuan (~$6,900) per DDR card, 70,000 yuan for the HBM version. Huawei targets production of 750,000 units in 2026.

The 950PR is the first Chinese AI chip gaining real orders from private-sector giants. If ByteDance and Alibaba actually switch from NVIDIA to Huawei — that’s a tectonic shift.

8. Wall Street Didn’t Reward NVIDIA for GTC — Stock Flat

Despite $1 trillion in projected compute demand and the Vera Rubin roadmap announcement, NVIDIA stock dropped during Huang’s keynote and ended the week essentially unchanged. A classic “buy the rumor, sell the news” move.

TechCrunch: “Why wasn’t Wall Street won over?” Goldman Sachs maintains a buy rating, but investors weigh AI bubble risk — NVIDIA trades at 38x earnings, with no margin for error.

9. Trump: National AI Policy Framework — Federal Primacy Over States

On March 24, the White House published the National AI Policy Framework — a document unifying AI policy at the federal level. Key goals: accelerate innovation, protect families, develop the workforce.

Critical: the framework gives federal regulations priority over state laws in areas of “national interest.” DOJ continues preparing to challenge state AI laws deemed overly restrictive.

10. Meta Lays Off Several Hundred — Reality Labs, Recruiting, Sales

On March 25, Meta confirmed laying off several hundred employees (under 1,000) — cuts hit Reality Labs, social media teams, and recruiting. This is Meta’s second round of 2026 layoffs (after January’s 10% Reality Labs reduction).

Context: Meta projects 2026 expenses of $162–169 billion, mostly on AI infrastructure. Simultaneously, the company spends millions recruiting top AI researchers.

11. OpenClaw — Three Releases in One Day, Fastest-Growing Open Source Project

On March 25, OpenClaw shipped three updates in a single day — the AI agent framework is growing at an unprecedented rate on GitHub. After the NVIDIA partnership announcement at GTC and the NemoClaw launch, the community exploded.

OpenClaw lets you launch an AI agent with one command — no Python, no graphs, no chains. Configuration via a SOUL.md file. Works on any OS, integrates with 50+ platforms.

12. Pentagon CTO Michael’s Conflict of Interest in Anthropic Case

Axios revealed that Emil Michael — the Pentagon CTO who pushed Anthropic’s supply chain risk classification — holds financial stakes in Anthropic’s competitors. Judge Lin referenced this in her ruling.

Michael led negotiations with Anthropic and personally signed the supply chain risk designation after talks collapsed. The Pentagon maintains there was no conflict of interest.

13. Microsoft Supports Anthropic in Court — Files Its Own Amicus Brief

Microsoft filed its own amicus brief supporting Anthropic in the Pentagon dispute — separate from the brief by 30+ OpenAI and Google employees. This matters because Microsoft is OpenAI’s primary investor and partner, yet defends its competitor.

Microsoft uses Claude in Copilot Cowork — it has a direct interest in Anthropic not being destroyed by the government.

14. State AI Regulation — ~200 Bills Tracked

By end of March 2026, Manatt Health tracks approximately 200 state-level AI bills. Key themes: mental health chatbots, patient consent, election deepfakes, AI developer liability.

California is processing SB 1159 (AI privacy) and AB 1898 (employer obligation to disclose AI workplace surveillance). Virginia sent a bill to the governor requiring labels on AI-generated political content.

15. DeepSeek V4 — Still No Official Launch

End of March 2026, and DeepSeek V4 still has no official launch. All previous windows — February, early March, post-Two Sessions — have passed. On March 18, Reuters discovered the anonymous “Hunter Alpha” model is actually Xiaomi’s MiMo-V2-Pro, not DeepSeek V4.

On March 9, something called “V4 Lite” appeared on DeepSeek’s website, but without official confirmation. DeepSeek’s API still doesn’t list a V4 model ID.

16. NVIDIA IGX Thor — Physical AI at the Edge Now Available

NVIDIA announced general availability of the IGX Thor platform — an industrial-grade platform for physical AI at the edge. IGX Thor enables real-time sensor data processing for robotics, manufacturing, and healthcare.

In healthcare, NVIDIA launched: Open-H (medical robotics dataset), Cosmos-H (synthetic data generation models), and GR00T-H (vision-language-action model for clinical tasks).

17. Amazon Orders 1 Million GPUs — Largest Chip Order in History

On the sidelines of GTC, it emerged that Amazon plans to purchase 1 million NVIDIA GPUs (plus other AI infrastructure) by end of 2027 for Amazon Web Services. This is the largest single AI chip order in history.

The order is part of a broader trend: Huang estimates total compute demand will exceed $1 trillion through 2027.

18. Healthcare AI Regulation — FDA Seeks Balance

Healthcare Brew (March 25): the FDA is developing a regulatory framework for medical AI designed to be flexible as the technology evolves rapidly. At HIMSS the previous week, the dominant question was: how to validate AI agents in hospitals?

The problem: current drug and device approval processes don’t fit AI models that update weekly. The FDA is exploring a “continuous oversight” model instead of one-time certification.

19. Amodei vs. Altman — AI’s Loudest Feud Continues

All of March has been an escalation of the Amodei-Altman dispute. Timeline: Amodei calls OpenAI’s messaging “lies” and “safety theater” → Slack messages leak showing Altman claiming he tried to save Anthropic → court blocks the ban → Pentagon announces appeal.

Axios: the case revealed Altman was “looking for an opening” when Anthropic was most vulnerable. Bloomberg: “Altman’s mishandled Pentagon deal works in Anthropic’s favor.”

20. Apple, Google and Siri — End of ChatGPT’s iPhone Monopoly

Apple’s announcement of opening Siri to external AI services is a strategic pivot. Until now, ChatGPT was the only assistant integrated with Siri. Starting with iOS 27, users will be able to route queries to Gemini, Claude, and others directly from Siri.

Apple positions itself as a neutral AI platform — not building its own model, but providing access to the best. For Anthropic (Claude) and Google (Gemini), this is a massive opportunity to reach hundreds of millions of new users.

---

Weekly Takeaway

The week of March 23–29, 2026 is the week of verdicts and collapses. A federal court blocks the Pentagon’s Anthropic ban in a landmark First Amendment ruling. OpenAI shuts down Sora — $15M/day costs against $2.1M in revenue — and loses a billion-dollar Disney deal. Trump appoints a tech council without Musk or Altman. Huawei’s 950PR wins orders from ByteDance and Alibaba — China’s chip ecosystem gains momentum. Apple opens Siri to the competition. And leaked Slack messages from OpenAI cast new light on the biggest feud in AI history.

---

Sources: Axios, CNBC, CNN, NPR, Washington Post, Bloomberg, Fortune, TechCrunch, Reuters, Variety, Deadline, Defense One, Breaking Defense, The Hill, Apple Insider, Healthcare Brew, Transparency Coalition, Sherwood News, NVIDIA Blog