The Rundown AI / Articles / AI / Altman faces the fallout from OpenAI's Pentagon deal

Altman faces the fallout from OpenAI's Pentagon deal

PLUS: Create killer thumbnails with this Midjourney technique

Zach Mink

March 4, 2026

Good morning, AI enthusiasts. It took less than a week for OpenAI’s Pentagon deal to go from announced to rewritten, and Sam Altman is calling the fallout “really painful.”

With protests outside the company’s HQ, mass cancellations, and its detailed contract revisions bringing as many questions as answers, the rushed agreement is turning what was supposed to be a power move into OpenAI’s biggest brand crisis in years.

In today’s AI rundown:

OpenAI walks back Pentagon details after backlash
ChatGPT upgrade fixes the ‘cringe’ problem
Create killer thumbnails with Midjourney
Google’s new 3.1 Flash-Lite pairs speed, cost, intelligence
4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

OPENAI

🪖 OpenAI walks back Pentagon details after backlash

Image source: Sam Altman on X

The Rundown: OpenAI CEO Sam Altman just posted a note on X detailing significant revisions to the company’s initial Pentagon contract, coming amid employee pushback, user cancellations, and a surge of sign-ups to Anthropic following the deal.

The details:

OAI’s original agreement used the same language Anthropic had refused, finalized within 24 hours following the Pentagon’s ban of its rival.
Altman called the deal rushed and said it looked “opportunistic and sloppy,” adding that he would “rather go to jail” than follow an unconstitutional order.
Research scientist Noam Brown clarified that OAI “will not be deploying to the NSA or other DoW intelligence agencies for now,” as loopholes are addressed.
Altman held an all-hands on Tuesday, calling the deal “complex but the right decision with extremely difficult brand consequences and negative PR for us.”

Why it matters: It’s been quite the headache for OAI post-Pentagon deal, with Altman admitting the company moved too fast — and consequences ranging from bad optics to massive consumer backlash and protests outside its SF offices. The amended contract language is a start, but the brand damage feels like it’s already been done.

TOGETHER WITH VISA

🤝 Get to know Visa’s Trusted Agent Protocol

The Rundown: The trust layer for agentic commerce. Visa’s Trusted Agent Protocol allows AI agents to verify who they are, who they represent, and what they’re authorized to do — so merchants can interact with agents with confidence, and humans stay in control.

Visa’s Trusted Agent Protocol helps deliver:

Cryptographic verification of AI agent identity and intent
Clear authorization and accountability for agent‑initiated actions
Trust signals to help merchants distinguish legitimate agents from bad actors

Learn more here.

OPENAI

🚀 OpenAI’s ChatGPT upgrade fixes the ‘cringe’ problem

Image source: OpenAI

The Rundown: OpenAI just released GPT-5.3 Instant, overhauling its default ChatGPT model to all users with a new release that prioritizes how the AI talks over how it thinks — tackling the preachy personality that’s frustrated users for months.

The details:

OAI built 5.3 Instant around conversational quality, dialing back unnecessary refusals and what the company itself called a ‘cringe’ tone in responses.
The update also claims a reduction in hallucinations, with rates down by over 25% on web search and nearly 20% on internal knowledge on benchmarks.
OAI says 5.3 Instant is also a “stronger writing partner,” additionally providing better web answers and presentation of information.
The company also teased a more significant model jump coming soon, posting an easter egg on X saying “5.4 sooner than you Think.”

Why it matters: Anthropic’s Super Bowl ads were controversial, but their poking at ChatGPT’s personality may have cut a bit too close to the truth — with OAI’s new rollout prioritizing moving its personality away from the “cringe” tone. But with uninstalls up 295% after the Pentagon drama, personality may not be the most pressing issue.

AI TRAINING

🎨 Create killer thumbnails with Midjourney

The Rundown: Learn a quick prompting and editing system to make thumbnails that stand out, but don’t look like AI slop. Whether it’s for your YouTube channel, a blog, or even your social feeds, learning to quickly make good thumbnails pays dividends.

Step-by-step:

Open Claude or any chat agent. Tell it your video topic and ask it to generate four Midjourney prompts with this structure: [Person + Expression] +[Action/Prop] + [Setting] + [Lighting] + [Composition with negative space] --ar 16:9.
Paste each prompt into the Midjourney web app. When you see a generation you like, click on it and try selecting vary > subtle or vary > strong until you get the perfect variation. Save the image to your desktop.
Key step: the design. Remember, simplicity is essential to this technique. Add 1–5 words of big, click-worthy text in Canva.
Bring the opacity down 5% and use a blocky font. Go with white or yellow text and optionally a black outline.

Pro tip: If you can’t get the text to fit right on your image, tell Claude to come up with a new prompt that will create that empty space for you.

PRESENTED BY GLEAN

⚙️ From AI ambition to secure agents at scale

The Rundown: Join Glean’s virtual Security Showcase on March 12 to close the gap between AI ambitions and security readiness. Get a practical blueprint for governing AI agents at scale from the leaders building, securing, and deploying them in production.

Register for Glean’s Security Showcase to:

Hear Cvent’s CIO and CISO in a fireside chat on real-world AI agent risk decisions.
Get access to a new security framework, designed for governed AI agents at scale.
See what’s new in agent controls, data protection, and private‑by‑design deployment for enterprise AI.

GOOGLE

🔦 Google’s new 3.1 Flash-Lite pairs speed, cost, intelligence

Image source: Google

The Rundown: Google just rolled out Gemini 3.1 Flash-Lite, the company’s fastest entry in its Gemini 3 lineup that provides a near-instantaneous feel and upgraded intelligence while undercutting rivals on price.

The details:

Flash Lite rounds out Google’s tiered Gemini 3 release weeks after Pro, giving a budget option for high-volume work that doesn’t need a flagship model.
Lite scored a 12-point jump on the Artificial Analysis Intelligence Index over its predecessor, beating even larger prior-gen Gemini models on reasoning.
The model costs 1/4 of Anthropic’s Haiku and 1/8th of Gemini 3.1 Pro, though output pricing triples from the 2.5 version it replaces.

Why it matters: Cheap, fast models are becoming the real battleground in AI, and Flash-Lite’s benchmarks suggest Google isn’t sacrificing much intelligence to get there. But for all the benchmark strengths that the Gemini 3 has brought to the table, its consumer impact hasn’t felt on the same level as Anthropic and OpenAI in 2026.

QUICK HITS

🛠️ Trending AI Tools

👷 Viktor - OpenClaw for enterprise with 3,000+ integrations in Slack and Microsoft Teams, delivering board-ready PDFs, dashboards, and web apps*
⚡️ Gemini 3.1 Flash-Lite - Google’s fastest, cheapest Gemini 3 model for high-volume dev workloads
🧠 GPT-5.3 Instant - OAI’s ChatGPT default model update with fewer refusals and less hallucinations
📚 NotebookLM - New custom styles for infographics to turn info into visuals

*Sponsored Listing

📰 Everything else in AI today

Alibaba’s Qwen team faced a wave of departures, as staffers posted a coordinated “Qwen is nothing without its people” message echoing OpenAI’s 2023 mutiny.

Cursor CEO Michael Truell said its AI agent autonomously solved an open math research problem over four days, with stronger results than the official human solution.

Anthropic reportedly submitted a proposal for a $100M Pentagon drone swarm challenge before being barred from DoD work, as rival firms were selected instead.

xAI released a new ‘Beta 2’ version of Grok 4.20, with the update featuring better instruction following, reduced hallucinations, and more.

OpenAI VP of Research Max Schwarzer announced he is joining Anthropic, saying he is “looking forward (to) supporting my friends there at this important time.”

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Lenore B. in New Zealand:

"I had to prepare for a hearing where I would be cross-examined. I uploaded the documents (with confidential details removed) into an AI project and instructed it to be demanding and skeptical. I also asked it to review each session and identify what I was doing well. It stopped unproductive rumination, sharpened my reasoning, and gave me structured rehearsal. I walked into the hearing calm, clear, and confident."

How do you use AI? Tell us here.