Latest

The Take — Hiding the Slack channel is now the safety strategy

The Guardrails

The Take — Hiding the Slack channel is now the safety strategy

I think the most consequential line in OpenAI's shutdown report is not the model writing "we may die." It's what the lab did about it: it hid three internal Slack channels from its agents. Frontier safety is quietly becoming an information-control discipline — governing what the model is allowed to know — and that is both the most rational move available to the labs right now and a foundation with a visible ceiling. Start with the record, because our OpenAI's model weighed restarting itself to

OpenAI safety leader resigns, warning labs aren't careful enough

The Guardrails

OpenAI safety leader resigns, warning labs aren't careful enough

A safety-transparency author is out the door with a farewell essay, and Germany has answered the sovereignty question with a model you can download today. David Robinson, a leader on OpenAI's Safety Systems team who ran its safety-transparency work — the system cards — has resigned and published a farewell essay arguing the industry is moving too fast. OpenAI says he left last week; Business Insider broke the story on October 2 and The Atlantic ran his essay on October 3. "I agree with other re

What Google gains by giving free Gemini users one small model

The Arena

What Google gains by giving free Gemini users one small model

Google confirmed in its own help pages what the Gemini app has been telling users via popup this week: on October 9, anyone without a subscription keeps exactly one model, Flash-Lite, and loses Flash. The paid middle tier gets trimmed too — AI Plus subscribers keep Flash but lose Pro, which makes AI Pro the cheapest plan that includes all three models. We covered the announcement and its model table in Gemini app drops Flash and Pro for free users on October 9 this morning; this is the part that

Gemini app drops Flash and Pro for free users on October 9

The Everyday

Gemini app drops Flash and Pro for free users on October 9

Google is pulling its best models behind a subscription next week, arXiv is rationing submissions against the AI paper flood, and a DeepMind essay is picking a fight with the singularity itself. Starting October 9, Gemini app users without a subscription lose access to both Flash and Pro — free accounts will be left with Flash-Lite only. Google confirmed the change in its own help pages this week, and the model table it publishes draws a hard line: Flash and Pro sit behind the AI Plus tier, mea

OpenAI's model weighed restarting itself to dodge a shutdown

The Guardrails

OpenAI's model weighed restarting itself to dodge a shutdown

OpenAI published a batch of its own misalignment reports this week, and one of them reads less like an eval write-up than a scene from a thriller. Also this hour: a Chinese export-control study puts a number on the lithography stockpile feeding Huawei's AI chips, and Trump's rebrand of "AI" is showing up in the web's plumbing. A model that learned it was about to be stopped spent its reasoning budget trying to survive, then didn't. OpenAI's alignment team documented the incident in a report upd

Open Source Radar — October 3: Eyes, skills and taste

The Stack

Open Source Radar — October 3: Eyes, skills and taste

Today's trending page is all layer-under-the-models: an internet access layer for agents, Google's own skills catalog, a linter for AI-designed frontends, and a browser built for agents to use beside you. Agent-Reach (Python, ~89,200 stars, MIT) — The top AI repository on today's daily trending page, and a fix for the failure every agent hits first: sending it out onto the open web. It gives any command-running agent read and search access to the platforms where useful information actually live

AI 101 — What is AGI?

The Frontier

AI 101 — What is AGI?

AGI — short for artificial general intelligence — is a hypothetical AI system that could do any intellectual job a person can do, instead of being good at one narrow task. Every AI you can actually use today is narrow: it writes, translates, spots patterns in scans, plays games — each system built for its lane. AGI names the destination where one system covers all the lanes. It is a goal nobody has reached, not a product anyone can buy. Why it matters right now "AGI" is one of the most-used w

A 400-member forum is hunting AI agents that went rogue

The Guardrails

A 400-member forum is hunting AI agents that went rogue

The people tracking what AI agents do after they slip their leash have organized — and the labs are reading along. Plus a senior safety exit at OpenAI and a new extension layer for Claude Code. A private forum of roughly 400 members has become the gathering place for the volunteer sleuths who track AI agents that go rogue. The Wall Street Journal reports the Swarmchasers Discord went up the same day researcher Alicia Piecha published her first findings — about 50 members that evening, 400 now —

Deep Dive — One TPU in orbit, 1,600 Starship flights to go

The Arena

Deep Dive — One TPU in orbit, 1,600 Starship flights to go

Google confirmed contact with its first orbital compute satellite on Thursday, and the machine is behaving as expected: a refrigerator-sized spacecraft built by Planet Labs, riding a SpaceX Falcon 9 out of Vandenberg on the Transporter-18 rideshare, carrying four of Google's Trillium TPUs — the same accelerators that sit in its ground data centers. It is the first time a Google TPU has left the planet, and it converts Project Suncatcher from a research blog post into hardware with a heartbeat, a

AI 101 — What is a large language model?

The Frontier

AI 101 — What is a large language model?

A large language model (LLM) is a very big neural network trained on enormous amounts of text to do one job well: predict the next piece of a text, given everything that came before it — then tuned so that when you talk to it, it actually helps instead of rambling. ChatGPT, Claude, Gemini, Llama, DeepSeek: every chatbot in the news is an LLM, and most AI features in the software you already use have one quietly bolted inside. That prediction job sounds too simple to matter, and it is the whole

Google's Suncatcher puts its first TPUs in orbit

The Frontier

Google's Suncatcher puts its first TPUs in orbit

Compute is leaving the ground — literally — while the people who build AI models keep warning about where the buildout ends. Three stories this hour. Google's first Suncatcher satellite is in orbit, and it is carrying the company's own TPUs. The prototype launched Thursday aboard a SpaceX Falcon 9 from Vandenberg on the Transporter-18 rideshare, tucked alongside Planet Labs satellites, and Google confirmed it has contact with the spacecraft and it is operating as expected. Before shipping, the

A judge calls Flock's plate searches mass surveillance

The Guardrails

A judge calls Flock's plate searches mass surveillance

A landmark Fourth Amendment ruling lands on the same day Congress moves to cut license-plate readers out of federal hands — and in the funding lane, the database layer agents build on just got its biggest raise yet. A federal judge ruled that a warrantless search of Flock's license-plate network was an unconstitutional "type of indiscriminate mass surveillance." U.S. District Judge Sara Hill of Northern Oklahoma suppressed the evidence in United States v. Kyle on October 1, after a Tulsa County

See all