Welcome
Who can say what AI can and cannot do?
This week’s center of gravity was accountability – who can say what AI can and cannot do. A federal judge called the Pentagon’s blacklisting of anthropologists a “disgrace.” "Attempt to paralyze." Two separate juries found Meta liable for designing addictive platforms. OpenAI quietly killed Sora, admitting that it was spending $15 million a day on a product that no one was committed to.
Amnesty International Weekly
Shepherd
Become an AI consultant and deliver “ROI with AI” to your clients
AI is transforming every workplace — but executives fear becoming one of those companies that “doesn’t get a return on AI investment.”
This is where you come in, and how you can Build 6-figure consulting By innovating using proven AI methods to achieve rapid ROI in AI projects.
Click here to request access to the AI Consulting Project →
In the news
Artificial Intelligence Weekly Briefing:
March 20-26, 2026
OpenAI kills Sora, a judge questions the Pentagon, and a jury decides that addiction by design is now a legal liability.
🎬 Watch and listen first
Jensen Huang: "I believe we have achieved Artificial General Intelligence (AGI)." March 22 Lex Friedman Podcast #494
→ The four most controversial words in AI this year. Huang redefines artificial general intelligence as "The artificial intelligence that builds a multi-billion dollar company" – Conveniently requires Nvidia chipset at every step.
Courtroom Week
Judge calls Pentagon humanitarian blacklist "Attempt to paralyze" March 24 CNBC
→ The strongest judicial response yet against the official’s AI campaign. If the injunction is issued, the moral red lines become legally defensible — and the Pentagon’s post-political approach to social media may be dead on arrival.
Meta and YouTube were found liable in Landmark’s social media addiction trial March 25 NPR
→ Jury in two days – $6 million in Los Angeles plus $375 million in New Mexico. The legal strategy of targeting design choices rather than content has led to 2,000 pending cases. This is Big Tobacco’s Big Tech moment.
Silicon transformations
Arm ships its first-ever chip — ending 35 years of pure licensing March 24 CNBC
→ Developed in collaboration with Meta and already ordered by OpenAI and Cloudflare, the AGI CPU features double the performance per rack of x86. Arm expects $15 billion in annual revenue from the chip by 2031. The stock jumped 16%.
Reflection AI eyes $25B valuation on $2.5B raise backed by Nvidia March 25 Reuters
→ A subsidiary of DeepMind is less than two years old, with minimal revenue, tripling its value to $25 billion. the "Sovereign artificial intelligence" His doctoral thesis – Open Source Models for Allied Nations – is now a $2.5 billion bet.
OpenAI Spring Cleaning
OpenAI kills Sora, and Disney’s $1 billion deal collapses March 24 Hollywood Reporter
→ Sora burned through roughly $15 million per day in inference costs for $2.1 million in lifetime revenue. Disney was told 30 minutes into the business meeting that the product had been discontinued. Fidji Simo of OpenAI called it “ "Side quest" – Now the company is moving aggressively towards projects and an IPO in the fourth quarter.
Robots walking the red carpet
Melania Trump enters the White House with a human-like robot March 25 CNN
→ Figure 03, greeting world leaders in 11 languages at an education summit, posed for a surreal photo. But the real story is that the First Lady installed human-shaped teachers in every home, a shift in Overton’s robot policy that happened before most people noticed.
AI tax on workers
Oracle plans to lay off up to 30,000 employees to fund the construction of an artificial intelligence data center March 5 Bloomberg
→ The clearest example yet of AI capital spending pressure: cutting 18% of your workforce to fund data centers that won’t return cash until 2030. Oracle joins Block (4,000 cuts) on the growing list of companies where humans are funding the machines they replace.
The supply chain is the attack surface
LiteLLM backdoor via hacked Trivy Scanner – 3.4 million daily downloads exposed March 24 Snake
→ A security scanner was compromised, compromising the LLM gateway used by 36% of cloud environments. The credential-stealing malware was present for three hours. The TeamPCP threat group is now openly collaborating with $LAPSUS, calling for open source security status "joke."
In 2026, the question is no longer an issue "Can artificial intelligence do this?" that it "Who decides what to do, and who pays when things go wrong?"
Amnesty International Weekly
This RSS feed is published on
https://aiweekly.co/. You can also subscribe
Via email.
