Skip to content
Tomasus
Go back

AI Digest W24: Going Public, Holding Back

2 min read

Two threads ran through this week, and they pulled in opposite directions. The labs are racing to go public and ship their strongest models, while quietly building brakes into those same models so they refuse to do the most dangerous things. Power and restraint, announced in the same breath.

AI Digest hero

The money news came first. OpenAI confidentially filed for an IPO on Monday, a little over a week after Anthropic did the same. CNBC points out OpenAI expected the paperwork to leak, so it just announced the filing itself, while saying it has not picked a timing yet. Two of the biggest private companies on earth heading for the public markets within days of each other. The AI IPO pipeline is now worth trillions, and Wall Street is bracing for it.

The model story is more interesting to me. Anthropic released Claude Fable 5 to the public, a version of its Mythos-class model that scored more than 10% above Claude Opus 4.8 on some benchmarks. The catch is built in: in high-risk areas like cybersecurity, biology, and chemistry, Fable refuses and quietly falls back to the weaker Opus 4.8 instead. Simon Willison’s worry, in a sharp post, is that you might never know when this happens, you just get a worse answer with no warning. Nathan Lambert calls these the new safety fables, and reads the release as the moment safety policy started shaping what a frontier model will and won’t say for everyone.

Elsewhere, Apple finally showed its hand. At WWDC it unveiled a rebuilt Siri that can hold a real back-and-forth conversation, running on a custom Gemini-derived model from Google rather than something Apple trained alone. Admitting you need a rival’s model to fix your assistant is a quiet kind of honesty. On the enterprise side, Anthropic tapped TCS to push its models into big companies at scale, the unglamorous plumbing that makes a near-trillion-dollar valuation real. And a new arXiv paper, “Agentic Software”, argues the shift is deeper than tooling: when the model is the reasoning engine and code becomes throwaway scaffolding, what software even is starts to change.

So the labs want to be public, powerful, and safe, all at once. Can you be all three? That tension is the story to watch.

T.


Share this post on:

About Tomasus

Someone who wants to understand what is coming and how it will impact us as human beings. Writing notes on AI, cybersecurity, history, and staying sane.


Series: AI Digest


Related Posts


Previous Post
AI Digest W25: When the Government Pulls a Model
Next Post
PAI in the Wild: Three Ways to Build Your Own