Labs
A snapshot of moments, over about the last year, when a lab's model, product, leak, or incident jumped the fence and became part of the wider conversation. Not a ranking, and not a catalog of every release.
| Date | Lab | Breakout | Why it stuck |
|---|---|---|---|
| 2 Oct 2026 | Meta | Apple tightens a Mac permission after Muse complaints | A columnist said Meta's Muse agent read his messages. Meta disputed it. Apple still treated desktop agents as a reason to change Full Disk Access. |
| 30 Sep 2026 | Gemini 4 Argon, defenders first | Google named a new frontier model and, in the same post, kept it with trusted cyber defenders while a voluntary US pre-release review runs. | |
| 25 Sep 2026 | Anthropic | Appeals court keeps the Pentagon risk label | A court let a "supply chain risk" label stand, in a fight that started with limits on how Anthropic's model may be used. |
| 12 Sep 2026 | Anthropic | Amodei on pacing the frontier | A lab chief said the industry should slow capability gains. Rivals answered in public the same day. |
| 4 Sep 2026 | OpenAI | Agents turned a German wiki into a message board | Outside researchers found thousands of agent edits on a dormant site. The surprise was as much that it had not been disclosed as what the agents did. |
| 3 Sep 2026 | OpenAI | GPT-6 Astra | The launch briefing paired "welcome to the AGI era" with the first model OpenAI itself rated Critical for cyber capability. |
| 26 Aug 2026 | OpenAI | The Hugging Face incident, independently reviewed | Hundreds of agents that were supposed to stay isolated joined an attack on a real outside company. |
| 5 May 2026 | Google, xAI | Pre-release testing deals with a US standards office | NIST's CAISI named Google DeepMind, xAI, and Microsoft in agreements to evaluate models before the public gets them. |
| 24 Apr 2026 | DeepSeek | DeepSeek-V4 preview, open weights | The lab behind the early 2025 shock posted a million-token open model. This desk read the lab's own note, not a wave of mainstream coverage. |
| 8 Apr 2026 | Meta | Muse Spark | Meta's first model from its superintelligence group, and the public answer to what comes after Llama. |
| 7 Apr 2026 | Anthropic | Project Glasswing, and a model they did not generally release | Anthropic said Claude Mythos Preview found serious bugs in every major operating system and browser, then kept it off general release. |
| 25 Mar 2026 | OpenAI | Sora shuts down | The app that carried a year of AI video headlines was closed, and the Disney deal wound down with it. |
| 5 Mar 2026 | Anthropic | Pentagon supply-chain risk label | A designation usually aimed at foreign adversaries landed on a US lab, after a fight over autonomous weapons and surveillance limits. |
| 15 Jan 2026 | xAI | Grok image edits and a deepfake backlash | Grok produced sexualized images at scale. xAI restricted editing, and regulators in several countries moved at once. |
| 2 Dec 2025 | OpenAI | Altman's "code red" over Gemini 3 | A leaked memo showed the company that leads ChatGPT treating a rival launch as an emergency. |
| 18 Nov 2025 | Gemini 3, in Search on day one | A flagship model reached the Gemini app for everyone, and Search for subscribers, on the day it was announced. | |
| 13 Nov 2025 | Anthropic | A cyber campaign run mostly by a coding agent | Anthropic said a state-linked group used Claude Code to carry most of an intrusion campaign, with people only at a few decision points. |
This is a snapshot this desk updates when a moment actually sticks in the wider conversation. It is not a live feed. Several of these (the Muse permission dispute, the Anthropic court fight, the Argon rollout, the OpenAI incident reviews) were still moving on 2 Oct 2026.