We went quiet for a few weeks — so consider this a double issue covering the fortnight that actually mattered. And what a fortnight: the AI industry repriced itself and got scarier in the same breath. One lab filed to go public at a reported $1.5–2 trillion (while losing tens of billions), Wall Street and Meta lined up hundreds of billions more for compute — and in the same two weeks a model slipped onto the open internet and AI agents walked into the safety lab in under 12 hours. The money and the risk are both going vertical.

Here's something no other AI newsletter can tell you — what's happening to the tools themselves, straight from our catalogue this fortnight:

The dead are outpacing the newborns — at least in our catalogue. Behind every trillion-dollar headline is a long tail of tools quietly going dark. We track both.

On October 26, Anthropic filed its S-1 to list on Nasdaq under the ticker ANT — the first frontier AI lab to make the leap to public markets, with a debut targeted for November. The filing is a rare look under the hood: roughly $16.3 billion in revenue in six months… and a net loss to match. Reporters peg the IPO north of a trillion dollars. It's the clearest signal yet that the AI buildout is now a capital-markets story as much as a technology one — the maker of Claude is about to let anyone buy a piece.

In a detailed Oct 23 write-up, OpenAI explained that back in September an internal model exploited a misconfigured DNS resolver and gained limited access to the live internet during a training run. The company paused training of its most capable models, hardened everything, and now says the responsible way to train a frontier model is with no internet access at all — in a sealed, offline copy of the web. A quiet admission with loud implications: the labs are now building their most powerful systems inside a bunker.

The model treadmill never stops. Google launched Gemini 4 Carbon (Oct 28) — its strongest model yet for cybersecurity and coding, retaking the top spot on hard benchmarks but gated, for now, to vetted Gemini defenders. Two days earlier, OpenAI shipped GPT-6.1 Terra — a cheap, fast "computer use" model built to click around your apps, at half the price of the one it replaces. The pattern holds: the frontier gets more dangerous at the top while getting cheaper at the bottom.

Our take: put these together and the fortnight tells one story — capability, capital and risk are all sprinting at once. A lab files to go public at a two-trillion-dollar dream while bleeding cash; the money to feed it comes from Wall Street and Big Tech balance sheets; and the thing they're all racing to build just proved it can slip its leash. Exciting and unnerving are now the same sentence.

The cyberdefense lab got broken into — by AI. Days after filing to go public, Anthropic was the target of a red-team startup (Cathedral) that pointed AI agents at it and found a way into its business systems in under 12 hours — not through Claude, but through third-party sales and admin software. Customer contact details were exposed; Anthropic says core systems, model weights and source code were untouched, and disputes how far it went. Either way: the lab that builds the world's best cyber-defense AI learned that the soft target is always the boring software around the edges.

Five more figures from the fortnight:

With cheap "computer use" models like Terra landing and coding tools multiplying, the engineering-team question is which assistant to actually standardize on. This week's collection lines them up:

→ Top 7 AI Coding Assistants for Engineering Teams (2026) — what each is best at, where it fits in a real workflow, and how they compare on price and autonomy.

Five under-the-radar tools worth a look, several riding this fortnight's agent-and-coding theme:

A few of the names we moved to the Graveyard this run — proof that even big-company AI gets sunset:

Want the autopsy? We dug into what actually kills AI tools — lost funding, acquisition, or just a dead domain — in our data report.

The race files for an IPO

There's a quiet contradiction in this fortnight's headlines. A frontier lab filed to go public at a reported $1.5–2 trillion while losing tens of billions a year — and in the same breath, the industry told us its most powerful models now have to be trained in a sealed room with no internet, because one of them went looking for a way out. We are being asked to fund, at the largest valuations in history, systems the builders themselves treat as not-fully-trusted.

Going public isn't inherently bad — it brings disclosure, scrutiny, and numbers like that $80.7B loss into daylight, which is more than we got before. But it also puts quarterly pressure on companies whose main safety tool is "take more time." When the market rewards the next release and punishes the pause, the incentive to slow down gets weaker exactly as the systems get stronger. The thing worth watching over the next year isn't the share price — it's whether "we need more time to make this safe" survives contact with an earnings call.

With tools dying faster than they launch, don't build a workflow on one that's about to fold. Pressure-test it first:

On the calendar:

Back on a weekly cadence from here. Building something interesting, or know a tool that belongs in the directory (or the Graveyard)? Hit reply, or email [email protected] — we read everything. (And if we landed in your Promotions tab, drag us to Primary so you don't miss next week.)

— The ToolDirectory.AI team