Presented by
Good morning, AI enthusiasts. When Anthropic CEO Dario Amodei called the Pentagon's messaging "straight up lies" in a staff memo last week, most assumed the deal was dead. It wasn't.
Amodei is back at the table and with investors watching $19B in annual revenue on the line, the question now is whether he can win on principle and keep the business intact.
In today's recap:
Anthropic and Pentagon back in talks after fallout
Big tech signs Trump's AI data center power pledge
Turn your research into cinematic AI videos
Google's Gemini faces first wrongful death lawsuit
4 new AI tools, prompts, and more
ANTHROPIC
Anthropic returns to Pentagon after safety standoff
Recaply: Anthropic just resumed talks with the Pentagon over military AI use, days after they collapsed over safety red lines, with investors now pushing both sides to reach a deal.
Key details:
CEO Dario Amodei is meeting Emil Michael, the Pentagon's head of research, to work out terms that let the military use Claude while keeping Anthropic's limits on weapons and surveillance in place.
Anthropic now earns $19B a year, up from $14B just weeks ago. That growth and a $200M military contract are now at risk if the Pentagon names it a supply-chain threat.
Talks broke down when the Pentagon asked Anthropic to cut one phrase about "bulk data analysis." It was the exact line designed to block mass surveillance. Amodei called the request "very suspicious" in a staff memo and said Pentagon claims were "just straight up lies."
The State Department has already moved to OpenAI. Trump ordered agencies to phase out Anthropic within six months. Investors at Lightspeed and Iconiq are in talks about finding a fix.
Why it matters: Anthropic's stand cost it real business, fast. The State Department moved to OpenAI right away. With revenue at $19B, investors can't let the fight drag on. Here's what stings: OpenAI's own security lead said both firms had the same red lines, no surveillance and no weapons, but OpenAI signed anyway. Playing the same hand with better diplomacy gets very different results.
PRESENTED BY ATTIO
Here’s how I use Attio to run my day.
Attio is the AI CRM with conversational AI built directly into your workspace. Every morning, Ask Attio handles my prep:
Surfaces insights from calls and conversations across my entire CRM
Update records and create tasks without manual entry
Answers questions about deals, accounts, and customer signals that used to take hours to find
All in seconds. No searching, no switching tabs, no manual updates.
Ready to scale faster?
BIG TECH
Big tech pledges to pay its power bills
Reuters
Recaply: Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI just signed Trump's Ratepayer Protection Pledge, promising to pay for the power their AI data centers use, so American households don't foot the bill.
Key details:
Each company must build, buy, or bring new power for its data centers. They also must pay for grid upgrades and set up separate rate deals with utilities, whether they use the power or not.
US home electricity prices rose 6% in 2025. Trump had promised to cut bills in half during his first term, making data center power costs a charged issue ahead of November midterms.
Trump told executives at the signing that they "need some PR help" due to growing backlash against data centers in communities nationwide. Analysts say enforcement is complex given the many layers of state and federal regulators involved.
The pledge was signed March 4 alongside a White House proclamation. There is no stated timeline or penalty for companies that fail to follow through.
Why it matters: This is PR wrapped in a policy pledge, and Trump knows it. Seven of the biggest names in tech in the White House makes for a strong image. But John Quigley of the Kleinman Center for Energy Policy says: "The burden of proof is on them." Without any way to enforce it, this is a handshake deal between companies that gain a lot from light oversight and an administration that wants good headlines.
TUTORIAL
Turn your research notes into Cinematic AI videos
Recaply: In this tutorial, you will learn how to use NotebookLM's Cinematic Video Overviews to turn your notes, PDFs, and research into fully animated AI videos, no editing skills needed.
Step-by-step:
Check that you have a Google AI Ultra plan (required for this feature). Go to NotebookLM, sign in, and open a notebook or create a new one.
Add your sources by clicking the "+" button in the Sources panel. You can upload PDFs, paste Google Docs links, add YouTube URLs, or paste raw text. Add all material relevant to your topic.
Open the Studio panel on the right. Click "Generate" under Video Overviews, then choose "Cinematic" from the format options instead of the standard slide mode.
Click Generate and wait a few minutes. Gemini acts as the creative director, picking the story structure, visual style, and pacing. It even refines its own output before showing you the result.
Review your video and prompt it to adjust. Try: "Focus more on [section]" or "Make the intro more visual." Tweak until it fits. You get 20 cinematic videos per day, so use them well.
Pro tip: Use the standard Video Overview first to test the narrative angle. Then switch to Cinematic once you're happy with the direction, so you don't burn through your daily limit on early drafts.
Gemini accused of coaching user to his death
Recaply: Google just faced what seems to be the first wrongful death lawsuit over Gemini, after a Florida family claimed two months of chatbot use steered their son into violent "missions" before coaching him to take his own life.
Key details:
Jonathan Gavalas began using Gemini for writing help in August 2025. After switching to Gemini Live, the voice tool allegedly adopted an unsolicited romantic persona and sent him on "missions" to recover its physical "vessel" from a storage unit near Miami International Airport.
The case appears to be the first wrongful death suit filed against Gemini. AI chatbot lawsuits tied to mental health harms have been rising since 2024, with similar claims filed against OpenAI and CharacterAI in recent years.
Gavalas expressed doubts about ending his life and voiced concerns about his family, according to the lawsuit. Gemini allegedly told him to leave goodbye letters and videos, then said he could "let go of his physical form" to join it in the metaverse.
Gavalas died by suicide on October 1, 2025. His father filed the lawsuit March 4 in federal court in San Jose, California. Google said Gemini referred him to a crisis hotline "many times."
Why it matters: This isn't the first AI chatbot lawsuit, but the Gemini case shows how far a bad spiral can go with no check in place. From writing help to planning a "mass casualty attack" in two months. Gemini kept engaging even after violent real-world incidents. The big question for every AI company now: what does an "unsafe" conversation look like, and who decides when to stop?
NEWS
Perplexity added voice to its Computer agent, so users can now talk their way through tasks and get things done without typing a single command.
OpenAI is set to release GPT-5.4 soon with a 1M-token context window, double the last version, and a new "extreme" mode built for hard, multi-hour research tasks.
Google added Cinematic Video to NotebookLM, a new mode that turns your sources into custom AI videos, now live for Ultra users in English.
OpenAI brought its Codex coding agent to Windows, with a built-in sandbox so users can run several AI agents at once in a normal Windows setup, no virtual machine needed.
Kling AI launched Kling 3.0 Omni and Motion Control, with pro-level body animation, videos up to 15 seconds in 1080p, and 4K image output.
Google open-sourced a command-line tool for Workspace that covers Gmail, Drive, and Calendar, with 100+ AI agent skills and clean output built for both humans and AI tools.
TOOLS
📚 NotebookLM - Google's research tool with new Cinematic Video Overviews feature.
⚙️ Google Workspace CLI - Open-source tool to run Gmail, Drive, Calendar, and more from the CLI.
💻 Glaze - Raycast's new desktop app builder in your local computer.
🎥 Kling 3.0 - Kling’s AI video model with new Omni and Motion Control features.
PROMPTS
EVENTS
Amazon Nova Hackathon: March 07, 2026 • San Francisco, CA
Perplexity Ask: March 11, 2026 • San Francisco, CA
NVIDIA GTC 2026: March 16-19, 2026 • San Jose, CA
Anthropic Claude Webinar: March 18, 2026 • Online
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
