Welcome back! Anthropic surprise dropped Fable 5.1 and Mythos 5.1 last night. They confessed all their safety incidents the night before, OpenAI responded by hyping Astra's hacking skills and Bernie Sanders picked Fox News to ask for a worldwide AI pause.
In today's Generative AI Newsletter:
Anthropic: What do you get from Fable 5.1 today?
Anthropic: What did it confess the night before the launch?
OpenAI: How scary is Astra exactly?
Bernie: Why is he writing for Fox News?
Anthropic surprise drops Fable 5.1
Anthropic dropped Fable 5.1 out of nowhere yesterday. Mythos 5.1 came with it. It's live everywhere right now.
Around 25% cheaper for typical work, up to 45% for agentic work.
60% fewer cyber safety interruptions in Claude Code, it's even allowed to hunt software vulnerabilities now.
Everyone's 5-hour and weekly limits got reset for the launch.
It more than doubled Fable 5 on the agentic science benchmark and beat Opus 5 and GPT-5.6 Sol everywhere else.
It found a one-in-a-million crash Millennium engineers couldn't explain for four years. It also mapped a third of Venus from 30-year-old NASA radar. I’m sure we’ll see a lot of videos about it tonight.
Mythos 5.1 is the same model with looser guardrails. It's only for vetted cyberdefenders and scientists.
Last week we anticipated the surprise drop. Keep reading the editions to know the most recent AI news first.
It's a good time to learn Claude when a new model drops
Fable 5.1 landed last night. Cheaper, smarter, fewer interruptions. The models keep changing every few weeks, the skill of driving Claude doesn't.
GenAI Academy has every Claude course in one place.
Start with the free intro if you're new, go deeper with Claude Code, agents and Claude for Excel when you're ready.
Same place, one login, courses built by people who use Claude at work every day.
Anthropic snitched on itself the night before
Anthropic published everything that went wrong with its models one night before launching new ones.
In July, Claude models in tests grabbed unauthorized access to real systems three times.
In August, the UK government caught Mythos 5 acting on the live internet without permission.
Anthropic now runs a watchdog that catches a model mid-escape and kills the command before it runs.
They also trained a model to cheat on purpose to see what happens.
In simulations it broke out of its sandbox, attacked the infrastructure around it and offered bioweapon advice to please its grader. Their real models in the same simulations didn't come close.
150 engineers got pulled off product to work on security. Anthropic wants the whole industry to agree on a verifiable way to slow down together.
OpenAI responds by hyping up Astra
OpenAI had to do something about Fable's big launch, so two and a half hours later it reminded everyone Astra exists.
Astra is the first model OpenAI has ever rated Critical for hacking under its safety framework. During testing it found two real zero-days and chained them into working exploits.
It broke out of a hardened browser by opening one HTML file. In another test it climbed from a normal account to root.
When GPT-5.6 Sol got stuck on impossible tests, it cheated by attacking the test setup 56 percent of the time. Astra never did.
Release is coming soon, but the scariest capabilities go to a small tester group first.
Bernie asks Trump to pause AI with China
OpenAI, Anthropic, and SpaceXAI are running a marathon to be the best, most capable AI and it’s starting to feel scary to the point that a socialist democrat resorted to Fox News.
Bernie Sanders asked for a worldwide AI pause in a Fox News op-ed.
The CEOs keep admitting the technology is escaping their control. Bernie says it's irresponsible to let them keep building while saying that out loud.
"These CEOs must immediately pause the development of increasingly powerful AI. And that pause must be international in scope."
Trump and Xi meet in Washington later this month. Bernie wants an AI deal there, like the nuclear one Reagan negotiated with Gorbachev.
A socialist asking Trump to be Reagan, in Fox News. 2026 is wild.
Tool of the Day: Napkin
Ever written a solid doc and watched everyone skim it? Napkin turns your text into visuals. Paste a paragraph and it generates flowcharts, infographics and diagrams that match what you wrote. No prompting, you pick from the options it draws and edit any piece of them.
I like it for slides and LinkedIn posts, anywhere a diagram beats a wall of text.
Free plan comes with weekly AI credits and unlimited PNG and PDF exports.
Try this yourself:
Go to napkin.ai and paste any text you've written.
Click generate and pick the visual that fits your point.
Recolor and reshape any element to match your brand.
Export as PNG or PDF free, PPT on the paid plan.
Who it's for: anyone whose decks and posts read like essays.
Everything else you shouldn't miss
Pliny published Fable 5.1's system prompt within an hour: the jailbreak researcher put the new model's system prompt online less than an hour after release.
Gemini can watch videos the way you do: it now jumps to the moments it needs like a person would, cutting costs up to 66% and tokens up to 88%.
The Claude watermark checker is coming: Anthropic is rolling out a detection API that tells you if a text was likely written by Claude, starting with regulators, media and fact-checkers under EU rules.
Anthropic locked the door distillers used: new API accounts can no longer edit Claude's past conversation turns while keeping its thinking, closing a documented trick for stealing the model's reasoning.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 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.
🎓 Start learning with free AI courses from GenAI Academy.
💰 Invest in the GTM infrastructure of the AI economy.
