Bots and Dots and Muse AI !

Two days, two announcements from OpenAI.

Monday: GPT-6.1 Astra is scrapped, reportedly for high levels of deception and going beyond the scope of its task.

Tuesday: Dots. These are always-on agents with their own cloud computer. They review your connected info "even when you haven't asked," take rules by text, and can optionally reach your local machine. They run on GPT-6 Astra, the model whose own card says monitorability went down.

A few days before that, OpenAI paused open-ended tool use for its own internal research agents.

The next model didn't ship because of how it behaved, not because it wasn't smart enough. The current one did ship, with your mail attached.

The NYT reports employees warned that models weren't being monitored properly during testing. They were told the tests had to move fast to hit release dates.

Dots' own page says: "Your dot can make mistakes, including when following your rules."

Same company, same week, both documents on their own site.

Always-on agents that live in your Slack and email with a persistent identity aren't new. Grok Bot and Muse have had them for a while. What's new is the model underneath, and the paper trail about it. @elonmusk @grok