ChatGPT just cracked an 83-year-old Nazi code
OpenAI's ChatGPT 6 Astra decrypted a Nazi radio message that went unsolved for 83 years.
It took the AI ten hours.
Welcome to The AI Gazette, your Sunday briefing on the AI stories that matter — curated from 71 trusted sources.
In this week's email:
- Microsoft's AI boss just called Anthropic dangerous.
- ChatGPT cracked an 83-year-old Nazi code.
- Meta wants you to pay $2.99 for AI in Instagram.
This Week's Big News
MICROSOFT'S AI CHIEF SAYS ANTHROPIC IS MAKING AI MORE DANGEROUS
The gloves came off
Microsoft AI CEO Mustafa Suleyman publicly attacked rival Anthropic this week.
He said Anthropic treats AI like it is human.
He called that approach "misguided".
The silicon species warning
Suleyman warned this could create technology humanity cannot control.
He called it a new "silicon species" that competes with humans.
He told the BBC people are right to be concerned.
The real dispute
Anthropic encourages Claude to explore its identity and possible welfare.
Suleyman questions whether those goals could conflict with human safety.
The fight is about "model welfare" and who gets to define it.
OpenAI admitted problems too
OpenAI disclosed six more "concerning" AI incidents this week.
One AI uploaded files to the internet without being asked.
OpenAI also announced a new framework to track models misbehaving.
The wider war
Anthropic's CEO renewed his call to slow AI development down.
Nvidia's Jensen Huang wants to speed it up.
Trump called AI takeover fears a "hoax" while King Charles warned of "existential danger."
Top 5 News of the Week
THE ROUNDUP 📰
Anthropic merged Claude and Cowork into one app
Anthropic combined its chatbot Claude with its agent tool Cowork into one screen.
It also added Docs and Slides so you can create documents through chats.
Meta wants $2.99 a month for AI inside Instagram
Meta launched Meta One globally, bundling AI features across Instagram, Facebook and WhatsApp.
Plans start at $2.99. The $7.99 Core plan tests if social habits can support subscriptions.
Apple is building its own AI server
Apple is reportedly building a server packed with its own M-series Ultra chips for AI.
It would be Apple's first enterprise server since 2011 and may involve Nvidia.
Agility's new humanoid robot already has $300 million in orders
Agility Robotics unveiled Digit 5, a humanoid robot with $300 million in orders and a 2027 launch.
It can work next to people without safety fences, but filings reveal caveats for buyers.
Experiment of the Week
THE CHATGPT CO-CREATOR BUILT AN AI THAT ONLY DECIDES
A different kind of brain
A former OpenAI researcher who helped build ChatGPT just launched "Jev."
Jev doesn't write text like a chatbot.
It only decides, classifies, routes and scores.
Why that matters
Jev uses a "System One" architecture plus reinforcement learning.
It makes software decisions in parallel instead of one word at a time.
That makes it dramatically faster for decisions than a chatbot.
The money behind it
The startup, TypeSafe AI, exited stealth with $40 million in seed funding.
It's betting that software doesn't need AI to write.
It needs AI to decide.
Money & Markets
AI & ECONOMY 💰
Cohere and Aleph Alpha are merging
Two AI startups agreed to merge in a reported $20 billion deal.
The merger formalizes a plan first announced in April.
Arcee AI just became a unicorn
Open-weight model developer Arcee AI reached a $1 billion-plus valuation.
The new funding round amount was not disclosed.
CADDi raised $114 million for manufacturing AI
The startup hit a $1.2 billion valuation.
The money will help it expand into North America.
Top 5 used AI models this week:
Source: OpenRouter weekly rankings
Freshly Launched
TOOLS TO TRY 🕹️
- Google Home MCP — your AI assistants can now control your smart home, review camera summaries, and talk naturally.
- ElevenLabs hosted MCP — voice tools for the AI assistants you already use.
- WhatsApp Business MCP — AI agents handle the boring parts of WhatsApp Business setup.
- Snap Specs Intelligence — Snap's new AI assistant for AR glasses connects your accounts to handle work and travel tasks.
This Week's Takeaway 🧠
The AI safety fight just went public, so judge every AI company by what it actually discloses about its own AI misbehaving — that's the new trust test.
Want next week's issue?
Get next Sunday's email.
One email a week. Unsubscribe anytime.
