AI
The most capable model on the market went offline three days after launch
Fable 5 arrived on June 9 with a million-token context window. On June 12, the government ordered it pulled.
The most capable model on the market lasted three days
Anthropic launched Fable 5 on June 9 and the government ordered it pulled on June 12. The same week, Tim Cook said goodbye with Siri running on a Google model, and OpenAI filed to go public with a projected $14 billion loss.
Business
Tim Cook says goodbye and Siri starts running on Google's model
In his final keynote as CEO, Apple showed an assistant rebuilt on a 1.2-trillion-parameter model.
From the studio
When the supplier disappears by someone else's decision
The week produced a case standard software architecture doesn't anticipate. A frontier model launched on Tuesday, with the largest context window on the market and availability across several clouds, and went offline on Friday by government order, alongside a sibling model.
Fallback chains were designed for a different kind of failure. They handle a server, a region or a version going down, because they switch address and carry on. When the entire supplier goes offline across every cloud at once, the only fallback that works is a different company, tested beforehand.
The week's other stories describe the same subject by different routes. A device maker decided not to build its own capacity and contracted a competitor's model for roughly a billion dollars a year. A customer who can't stop began testing a replacement for the model that vanished.
The question organising all of it is old and got more expensive to answer: when the preferred supplier doesn't deliver, what runs instead, and has anyone tested it under real load?
In this issue
Business
OpenAI files to go public with a projected $14 billion loss
The confidential filing went in on June 8. Profitability isn't expected before 2029.
Regulation
Pentagon tests replacing an AI model in classified systems
The evaluation involves models from two vendors as a possible swap for use in classified environments.
Security
Anthropic issues a rare warning about its own model
The company said its models may soon be too powerful to control.
Tools
A dependency firewall blocks malicious packages before they reach the requester
The tool reviews every open-source package downloaded in the company, including installs by AI agents.
Security
An enterprise AI security layer arrives with exfiltration detection
The package includes access control, multi-party approval and granular audit logging.
Regulation
A national security memorandum mandates accelerating AI adoption
The June 5 text requires adoption at scale with guaranteed reliability and resilience.
Tools
A corporate catalogue reaches 11,000 available models
The number describes the new problem: choosing, rather than getting access.
AI
A model ships with a million-token context and always-on reasoning
The large window changes what you can ask without preparing the material first.
Business
A billion-a-year deal puts a competitor's model inside the product
The estimated annual contract value describes the price of not building your own capacity.
Business
Two large technology listings in the same week
One rocket company priced and debuted; another filed a confidential registration.
AI
Five in-house models built jointly with a competing supplier
The new generation covers on-device and server execution, born from partnership rather than solo development.
Regulation
The European law's enforcement deadline enters its final countdown
Fifty-five days remained before requirements to document and publish a training-material summary.
AI
An extensions system puts multiple AI vendors on the same device
For the first time, iPhone users could choose between models from different companies.
Work
A leadership change at one of the largest technology companies
The chief executive announced his departure in April and handed over to the hardware chief in September.
Sections