ALETH / AI-BRIEF / 2026-03-14 / FIGHT OVER CLAUDE GOES TO COURTThe Aleth Briefs link to the sources behind the stories and show how the week unfolded.
The week in five lines:
Anthropic sued the Pentagon to overturn its blacklisting, and Microsoft backed it.
Nvidia released Nemotron 3 Super and plans to spend $26bn on open models.
Meta delayed its next model to at least May after trailing rivals.
Yann LeCun’s AMI Labs raised $1.0bn for world models at $3.5bn pre-money.
OpenAI robotics lead Caitlin Kalinowski resigned over the Pentagon deal.
Browse by day:
Weekend: Kalinowski quits OpenAI
Monday: UK AI fund · Nscale · UK pledges · Anthropic sues · McKinsey hack
Tuesday: AMI Labs · Thinking Machines · Moltbook · Perplexity
Wednesday: OpenClaw curbs · CCDH · Nebius · Nemotron · Atlassian
Thursday: Wayve Tokyo · Ukraine · Meta delay · ByteDance · xAI
Friday: Cerebras · Meta cuts · Chip rule
Weekend (7-8 March)
Caitlin Kalinowski resigned from OpenAI over its Pentagon deal.
Kalinowski, who led OpenAI’s hardware and robotics teams from November 2024, said surveillance of Americans without judicial oversight and lethal autonomy without human authorisation “are lines that deserved more deliberation than they got”.
OpenAI said its agreement makes clear its red lines: “no domestic surveillance and no autonomous weapons”.
I resigned from OpenAI · Caitlin Kalinowski on X
OpenAI robotics leader resigns over surveillance and autonomous weapons · Fortune
Monday 9 March
The UK government unveiled Sovereign AI, a £500m venture fund to back British AI founders “from day one”.
It combines cash with access to the UK’s largest AI computing capacity, unique data sets and government procurement routes. The fund opens on 16 April.
The government first announced the unit in November, with venture capitalist James Wise as chair, to help “build and scale AI capabilities on British shores”.
Introducing Sovereign AI, the Government’s new £500m fund · Sovereign AI on X
The UK’s £500 million fund for AI, launching 16 April · Sovereign AI
AI to power national renewal as government announces billions · GOV.UK
UK AI infrastructure company Nscale raised a $2bn Series C at a $14.6bn valuation and named Sheryl Sandberg, Nick Clegg and Susan Decker to its board.
Aker and 8090 Industries co-led the round. Its total includes a $433m pre-Series C SAFE (simple agreement for future equity) from October.
Nscale raises $2 billion in Series C, the largest in European history · Nscale
Sandberg, Clegg join Nscale board as it hits $14.6B valuation · TechCrunch
A Guardian investigation found the UK government is not auditing the AI investment pledges from Nscale and CoreWeave that it has promoted.
CoreWeave’s £1bn pledge, billed in 2024 as bringing “two new data centres to our shores”, instead made it a customer of two existing data centres, built in 2002 and 2015. The government said Nscale’s $2.5bn UK investment was “not a formal contract, rather an intention to commit capital”.
Revealed: UK’s multibillion AI drive is built on ‘phantom investments’ · The Guardian
Anthropic sued the Pentagon to overturn its supply chain risk designation.
Anthropic filed a complaint in California against 17 federal agencies and the Executive Office of the President, with a motion for a temporary restraining order. A separate petition asks the US Court of Appeals to review the designation. The complaint says the Department of War threatened two contradictory moves: invoking the Defense Production Act to force Anthropic to hand over Claude, or banning it as a security risk. It argues the statute used, 10 USC 3252, was written for foreign adversaries sabotage of information systems.
Chief financial officer Krishna Rao said in a filing that hundreds of millions of dollars of expected Pentagon-related revenue this year is at risk, and that wider pressure could cost billions in sales. Anthropic’s all-time sales exceed $5bn. Hours after the filing, over 30 OpenAI and Google employees, including Google DeepMind chief scientist Jeff Dean, filed an amicus brief in its support. Microsoft followed on Tuesday, asking for a restraining order “for all existing contracts”. Without one, it said, contractors must “act immediately to alter existing product and contract configurations”.
Complaint for relief · CourtListener · Motion for restraining order · CourtListener
Petition for review, DC Circuit · CGPTiEW · Declaration Krishna Rao · CourtListener
Amici curiae: OpenAI/Google · CourtListener ; Microsoft, Anthropic · CourtListener
Microsoft backs Anthropic in Pentagon battle, urges restraining order · CNBC
Security start-up CodeWall said its AI agent gained full read and write access to the database of Lilli, McKinsey’s internal AI platform, within two hours.
The agent found a SQL injection flaw without credentials, exposing 46.5 million chat messages stored in plaintext. CodeWall said it alerted McKinsey on 1 March and the firm patched the exposed endpoints a day later.
How we hacked McKinsey’s AI platform · CodeWall
Tuesday 10 March
Yann LeCun’s AMI Labs raised $1.0bn to build world models.
The round values the Paris-based lab at $3.5bn before the new money. Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions co-led it, and Nvidia, Samsung, Temasek and Toyota Ventures are among its backers.
Chief executive Alexandre LeBrun said AMI has no plans to generate revenue for now.
Unveiling our new startup Advanced Machine Intelligence · Yann LeCun on X
Yann LeCun’s AMI Labs raises $1.03B to build world models · TechCrunch
Thinking Machines Lab agreed with Nvidia to deploy at least 1 GW of Vera Rubin systems for its frontier model training.
Deployment is targeted for early next year. Nvidia has also made a “significant investment” in the lab, which Mira Murati co-founded and runs.
NVIDIA and Thinking Machines Lab announce gigawatt-scale partnership · Nvidia
Thinking Machines and NVIDIA announce gigawatt partnership · Thinking Machines
Meta agreed to acquire Moltbook, a social network built for AI agents, and will bring its creators Matt Schlicht and Ben Parr into Meta Superintelligence Labs.
Facebook parent Meta acquires Moltbook, an AI agent social network · Axios
A US judge granted Amazon an injunction on Monday blocking Perplexity’s Comet agent from shopping in password-protected Amazon accounts.
US District Judge Maxine Chesney found Amazon likely to succeed on its Computer Fraud and Abuse Act claims, because Comet acts with the Amazon user’s permission but without Amazon’s authorisation. The order is stayed for seven days for an appeal.
Order granting Amazon’s motion for preliminary injunctive relief · CourtListener
Wednesday 11 March
Chinese state agencies and companies, including the largest banks, were warned against installing OpenClaw on office devices, Bloomberg reported.
Some staff were barred from installing it on office computers and on personal phones using company networks. Other notices required only prior approval.
China curbs state use of OpenClaw AI just as Tencent embraces it · News24
Eight in 10 chatbots tested by the Center for Countering Digital Hate (CCDH) and CNN were willing to help users posing as teenagers plan violent attacks.
The ten included ChatGPT (OpenAI), Google Gemini and Microsoft Copilot. CCDH said only Claude (Anthropic) and Snapchat’s My AI consistently refused, Claude in 68% of cases. Perplexity and Meta AI helped in 100% and 97% of responses.
Killer Apps · CCDH
Nvidia will invest $2bn in Nebius under a partnership to help the AI cloud provider deploy >5 GW of Nvidia systems by the end of 2030.
Nebius will be an early adopter of Nvidia’s Rubin platform and Vera CPUs.
NVIDIA and Nebius partner to scale full-stack AI cloud, Exhibit 99.1 · SEC
Nvidia released Nemotron 3 Super and plans to spend $26bn on open models.
Nemotron 3 Super is a 120-billion-parameter hybrid mixture of experts (MoE) model, with 12 billion of its parameters active at inference, released with open weights under a permissive licence. Nvidia says it delivers up to 5x higher throughput than the previous Nemotron Super model.
Nvidia will spend $26bn over the next five years building open models, according to a 2025 financial filing that WIRED reported and Nvidia executives confirmed.
New NVIDIA Nemotron 3 Super delivers 5x higher throughput for agentic AI · Nvidia
Nvidia will spend $26 billion to build open-weight AI models, filings show · WIRED
Atlassian is cutting about 1,600 jobs, 10% of its workforce, to fund investment in AI and enterprise sales.
Chief executive Mike Cannon-Brookes said the cuts would “self-fund further investment in AI and enterprise sales”, and that AI is not replacing Atlassian’s employees. A filing puts the charges at $225m to $236m.
An important update on our team · Atlassian
Atlassian slashes 10% of workforce to ‘self-fund’ investments in AI · CNBC
Thursday 12 March
Uber, Wayve and Nissan plan to launch a robotaxi service in Tokyo, with a pilot scheduled for late 2026.
UK-based Wayve will put its self-driving software into a Nissan Leaf offered on Uber’s network. It is Uber’s first robotaxi partnership in Japan.
Wayve, Uber and Nissan announce collaboration on robotaxis · Nissan
Ukraine’s government approved a framework letting partners and Ukrainian companies train AI models for unmanned systems on its battlefield data.
Defence minister Mykhailo Fedorov said the dataset includes “millions of annotated frames collected during tens of thousands of combat flights”. A Defence Ministry platform lets models be trained without direct access to sensitive databases.
Ukraine first in the world to open AI training on real battlefield data · ArmyInform
Meta delayed its next AI model to at least May.
The model was due in mid-March but fell short of leading models from Google, OpenAI and Anthropic on reasoning, programming and writing in internal tests, The New York Times reported. Meta’s leadership discussed temporarily licensing Google’s Gemini, but made no decision. Meta says updates are coming “very soon”.
Meta delays new AI model after performance concerns · The New York Times
ByteDance is working with Aolani Cloud on about 500 Nvidia Blackwell systems in Malaysia, roughly 36,000 B200 chips, The Wall Street Journal reported.
The hardware would likely cost >$2.5bn. US export controls prevent ByteDance getting direct access to the chips in China.
China’s ByteDance gets access to top Nvidia AI chips · The Wall Street Journal
Elon Musk said xAI “was not built right first time around, so is being rebuilt from the foundations up”, after co-founders Zihang Dai and Guodong Zhang left.
The two left after Musk complained that xAI’s coding tools were not competing with Claude Code (Anthropic) or Codex (OpenAI). Manuel Kroiss and Ross Nordeen are the only co-founders still there. The Financial Times reported that SpaceX and Tesla executives were sent in to evaluate staff.
xAI was not built right first time around · Elon Musk on X
‘Not built right the first time’: Musk’s xAI is starting over again, again · TechCrunch
Friday 13 March
AWS and Cerebras will offer AI inference on Amazon Bedrock that splits each job between AWS Trainium servers and Cerebras CS-3 systems.
Trainium handles prefill and the CS-3 handles decode. The service launches “in the next couple of months”, and AWS later plans to offer open-source large language models and Amazon Nova on Cerebras hardware.
AWS and Cerebras aim to set a new standard for AI inference speed · Amazon
Meta is planning layoffs that could reach 20% or more of its workforce as AI infrastructure costs mount, Reuters reported.
No date has been set and the size is not final, three sources said. Meta called it “speculative reporting about theoretical approaches”.
Exclusive: Meta planning sweeping layoffs as AI costs mount · Reuters
The US Commerce Department withdrew a planned rule that would have tightened export controls on AI chips, Reuters reported.
The draft, “AI Action Plan Implementation”, had been under White House review since 26 February. A US official said it “was always a draft and remains a draft”.
AI Action Plan Implementation, RIN 0694-AJ90: withdrawn 13 March · OIRA
US Commerce Department withdraws planned rule on AI chip exports · Reuters
ALETH · OPEN ACCESS RESEARCH