Welcome back! OpenAI launched a cyber model that does exactly what Anthropic's Mythos was locked up for, and nobody moved to ban it. Google put $75 million into AI filmmaking and the artiest studio in film is making movies with AI now. Plus Trump's quantum order and 11 ways to stop hitting your Claude limit.
In today's Generative AI Newsletter:
Cyber: Why does OpenAI get to ship what Anthropic was banned for?
Hollywood: Why is the movie star suddenly optional?
Quantum: What's behind Trump's 2028 deadline?
Claude: How do you stop running out of Claude?
OpenAI shipped the full version of GPT-5.5-Cyber last night, a model built to hunt down software vulnerabilities and patch them, which Altman says posts state-of-the-art numbers on the CyberGym benchmark.
That is the exact capability Anthropic locked Mythos away for and got Fable 5 banned over.
Anthropic's pitch was that the capability was too dangerous to release. Researchers had already said it was nothing special, since the same vulnerability-hunting runs in GPT-5.5, Claude Opus and other models.
OpenAI gates its cyber model to vetted defenders, the same way Anthropic gated Mythos, but OpenAI's never drew a ban.
So why did one lab get banned and not the other?
First of all, it's not even a public release yet. And OpenAI is Washington's favorite child.
They hired a co-author of Trump's AI plan last week, while Anthropic spent the month fighting the same White House.
One lab is on Washington's good side, and it ships the same kind of model with none of the consequences.
Special highlight from our network
Bolting AI onto an existing workflow gets you a few wins. Redesigning the workflow around AI changes the economics. The gap between the two is widening every quarter.
Agency Advantage 2026 brings together agency leaders, AI practitioners, marketers, and operators for two days on what AI-native operations actually look like.
You'll learn:
• How leaders are redesigning workflows around AI
• How teams are scaling output without scaling headcount
• How AI agents are being deployed into real business operations
• What separates AI-assisted organizations from AI-native ones
June 30–July 1, 2026 • Free Virtual Event
Get good at AI this July
GenAI Academy is running three live cohorts, all starting July 6, all taught by people who actually do the work. Early-bird pricing closes this week.
Stop AI Slop: Train AI to write in your real voice with Chris Lele's Voiceprint Method, patterns instead of prompts. Two weeks, $197 with the 21% early-bird, 5 seats left.
AI Content & Monetization Engine: Build an AI content system that turns attention into revenue, with Lambros Lazopoulos, who has trained over 60,000 people. Four weeks, $398 at 20% off.
Data Storytelling with AI: Turn messy data into stories executives act on, taught by 13x Microsoft MVP Jordan Goldmeier. Three weeks, $900.
Google DeepMind is putting $75 million into A24, the studio behind "Everything Everywhere All at Once," and the two will build AI filmmaking tools together.
It's Google's first stake in a studio, and it's the one that stings, because A24 was the name everyone pointed to as the place that would never do this.
Except everyone's already in.
Netflix bought Ben Affleck's AI studio, Amazon's MGM built an AI unit and Lionsgate gave Runway its 20,000-title catalog to train on.
Now the prettiest holdout has joined them, and the whole business is wiring AI into how films get made.
The human actor turns optional once a model can generate a convincing performance.
Studios get to revive dead stars, de-age the living or build a face that never existed and never asks for a trailer. The same tech that erases the star could put you in the lead instead.
A24 now is helping build the machine that makes humans optional. If even A24 is in, the only question left is how fast the rest of it goes.
Trump signed an executive order telling the government to build its own quantum computer and put it in a Department of Energy lab, with his science chief betting it's doable by 2028.
It also tells agencies to quantum-proof their encryption before "Q-day," the point where a quantum machine could crack the codes guarding banks and state secrets.
Quantum has been ten years away for thirty years. Now Washington has put a 2028 clock on it.
If you caught yesterday's Atlas, you got the first ten ways to stop hitting your Claude limit, and you've been waiting on the rest.
We promised the other eleven. Here they are.
Most people burn through Claude by wasting it, re-uploading the same PDF, rambling in one endless thread or asking for the same summary twice, so these are the habits that stretch it.
11. Summarize and restart. Every 10 to 20 messages, ask for a summary, then paste it into a fresh chat.
12. Use Projects for repeat files. Upload that PDF once, not five times.
13. Turn features off. Keep Search, connectors and memory off until a task actually needs them.
14. New topic, new chat. Don't run a LinkedIn post, a proposal and a recipe in one thread.
15. Don't dump the whole folder. Point Claude at the specific files the task needs.
16. Schedule recurring tasks. Let the same weekly job run itself instead of redoing it by hand.
17. Use the right tool. Images go to ChatGPT, instant search to Grok, text to Claude.
18. Dictate your prompts. Whisper or any voice tool gives more context in less typing.
19. Set your preferences. In Settings, open General then Personal Preferences and switch off memory you don't need.
20. Give precise commands. "Make a bar chart from this CSV and save it as chart.png" beats "visualize this."
21. Spread usage across the day. Old sessions drop out of the rolling window, so don't burn it all in five hours.
Run these and most people get noticeably more done before they ever see a wall.
Most brand research lives in twelve browser tabs and a slide deck that's already out of date. Merciv pulls your internal data together with social chatter, syndicated research and the open web into one cited place, then hands back briefs and decks with the sources attached. It's a research analyst that shows its work.
Try this yourself:
Start free at merciv.com and ask it a plain-language question about your brand or category.
Point it at your own files, like reports, PDFs and old decks, so it answers from your data instead of guessing from the open web.
Have it run sentiment or a competitor breakdown across Reddit, TikTok, X and YouTube at once.
Export the answer as a cited deck or brief you can take straight to leadership.
Who it's for: brand, insights and strategy teams tired of buying a dashboard of mentions and assembling the story themselves.
SpaceX's supercomputer is for rent: SpaceX signed a deal worth up to $6.3 billion letting open-source startup Reflection AI run on its Colossus data center, at $150 million a month.
Japan found a loophole in the AI ban: Sakana's new Fugu model hits Fable and Mythos-level performance by orchestrating other models, so export controls have nothing to switch off.
Spy agencies say AI hacking is months away: The Five Eyes alliance put out a rare joint warning that frontier models like OpenAI's Daybreak and Anthropic's Fable 5 will supercharge cyberattacks within months.
Google opened its agent API to everyone: Its Interactions API is now the default way to build on Gemini, with managed agents and background tasks baked in.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
