TL;DR
The Pentagon just gave 3 million workers ChatGPT and Grok, and quietly left Claude off the list
Anthropic's Fable 5.1 found a bug that stumped engineers for five years, and cut agentic costs up to 45%
Webflow launched Source, letting marketers and AI agents build sites together
Meta shipped a voice model that transcribes, diarizes, and detects speech endpoints in one real-time pass
Monid gives your agent one key to access 1,800+ APIs and pay for tools on its own
Synthesia's new role-play sessions let you practice hard conversations with an AI that pushes back like a real person
Struggling to keep up with industry news while you build your empire? We were too! Until we found The CEO Report, a short end of day (tech-leaning) business news recap email, read by 250,000+ daily active CEOs/Founders/& Builders!
(it's the fastest way I've found to catch tech and business news without doom-scrolling for 20 minutes)
THE PENTAGON ROLLS OUT CHATGPT AND GROK 🤔
The Department of War announced Monday that ChatGPT and Grok are now live on GenAI.mil, its internal AI platform for military and civilian staff. That platform launched in December with just a military version of Gemini, and it already has 1.7 million users out of a 3 million person workforce. Add OpenAI and xAI to the list, and nearly the entire department now has government-approved versions of the biggest AI tools running in a secure environment, cleared for sensitive but unclassified work like HR, logistics, and contracting.
Funny enough, Claude isn't on the list.
Back in January, War Secretary Pete Hegseth publicly complained about AI models that "won't allow you to fight wars," a line later confirmed to be aimed at Anthropic, after the company tried to get contract language barring its tech from autonomous weapons or surveillance use. In February, Trump ordered federal agencies to stop using Anthropic altogether. Hegseth called the company a "supply-chain risk." Anthropic sued.
Last week, a judge ruled the Department of War acted unlawfully in that designation. Days later, ChatGPT and Grok get the rollout, and Claude is nowhere in the announcement.
OUR TAKE
Today’s story is about what happens when a company draws a line and a government client doesn't like where it lands.
Every other major lab said yes to defense work with no public friction. Anthropic asked for guardrails first and got benched for it, even after winning in court.
Worth watching whether this becomes the actual dividing line in the AI industry: labs that build for whoever pays, and labs that negotiate terms first. That split is going to matter more than any benchmark score.
The irony writes itself. The company that built its whole brand on AI safety just got locked out of the most safety-relevant deployment in the country.
QUICK HITS
Webflow gives marketers and agents the same desk to work at: Source by Webflow is a new platform (limited research preview) where marketing teams and AI agents build and optimize sites together on your existing codebase and martech stack. Agents get direct code access for bigger changes, humans get the visual canvas for the finishing touches, and every edit writes to the same code either way.
Meta ships a voice model that does three jobs most tools split across three: Muse Voice Transcribe handles real-time transcription, speaker diarization for 20+ people, and endpointing in one pass, and it's already topping streaming speech-to-text benchmarks. Live now in Meta AI for Mac and Muse Code, with API access at $0.18/hour.
Anthropic ships its strongest coding model yet, and makes it cheaper to run: Claude Fable 5.1 is generally available now, with Mythos 5.1 (same model, fewer guardrails) gated for cyber and life-sciences work. Pricing holds at $10/M input and $50/M output, but cache reads drop 75% to $0.25/M, cutting typical costs 25% and heavy agentic workloads up to 45%. Big gains showed up in multi-hour coding sessions and scientific research benchmarks, plus new Enterprise Frontier Safeguards giving companies more control over their own data.
Looking for the best way to reclaim your health? Factor offers 100+ weekly dietitian-approved meals, including including High Protein, GLP-1 Support, Calorie Smart, and Keto options.
PRACTICE THE HARD CONVOS…
Use Synthesia's new role-play sessions to rehearse tough conversations with an AI that pushes back like a real person, then get coached on exactly what to fix.
⏱️ 15 min | 🔧 Synthesia account (free tier works)
Why build this? You can watch every training video in the world and still freeze the second someone pushes back on you in real life. Performance reviews, sales objections, an escalating customer. Knowing the playbook and executing it live under pressure are two different skills, and most companies only ever train the first one. Synthesia's role-play sessions fix that by giving you an actual conversation to mess up in private, then telling you exactly where it went sideways.
Steps:
Head to the role-play sessions page and pick a scenario. Options right now: manager feedback, discovery call, cold call, retail sales.
Choose a persona. Each one changes the whole conversation. In manager feedback you can talk to an ambitious high performer wondering why she's not promoted, an underperformer, or someone checked out and disengaged.
Read your context card before you start. It tells you who you're talking to, what the conversation is about, and what your goal is, plus a brief on the person's development areas.
Turn on your mic and have the actual conversation. The avatar reacts in real time. Vague feedback gets pushed on. Bad tone gets called out. It's not a script.
Get scored and coached. You'll get a pass/fail score, a skill breakdown, and specific notes on what worked and what to try differently.
Retry immediately. Same scenario, same persona, coaching still fresh. This is where the score jumps.
Expected outcome: You walk out having actually practiced the conversation instead of just watching someone else have it. Managers and teams get visibility into who's improving and where the real skill gaps are, so training finally measures whether someone can handle the moment, not just whether they clicked through a module.
Full tutorial:
TOOL OF THE DAY
One API key gives your agent access to 1,800+ tools instead of you signing up, testing, and billing for each one separately. Your agent can search the catalog and pick a tool by price, reliability, or performance at runtime, so it's not hardcoded to call the same API every time. That fits how agent workflows actually behave, since the right tool for a task isn't always known ahead of time.
Three ways to connect, same registry and balance either way: Skill (one line into your agent's chat), MCP (add the remote server), or CLI (install and run from your terminal).
