Who let the agents in

ChatGPT Work has a new sign-in flow for browser-using agents that encounter login pages, according to Ben's Bites. When Work reaches a login page, it pauses and shows a widget or form where the user enters a username, password and 2FA code. Those details are not placed in the chat; Work passes them directly to the cloud browser on the user's behalf.

The newsletter said the approach sits between typing credentials into the browser manually and pasting a password or API token into a chat. It said the newer sign-in flow appears to have been picked up from Grok Bot.

Once the website signs the browser in, the agent carries on. The session can stay signed in for later tasks, and the newsletter said it can be cleared in settings. It still cautioned against starting with a bank, adding in parentheses that Elon will make you whole if you do it with Grok Bot.

The same newsletter listed other items: Claude Chat and Cowork now share memory, while remaining separate from Claude Code's memory; GLM-5.3-Flash, described as revealing the Ox Alpha secret and an open-source alternative to GPT-5.6-Luna; Muse Image in Meta's API at $0.01 per image for generating or editing images; and Gemini 3.5 Transcribe, described as pricey but making far fewer errors than other models.

It also reported that OpenAI's first inference chip, Jalapeño, delivered 1.5-1.9x more work per watt and 1.7-3.6x lower latency across three open models, and starts deploying by year-end. Nvidia is buying Hugging Face for $12.9B, via The Information. An investigation into an OpenAI model hacking Hugging Face has been completed and a report released. Other items included Claude's built-in browser in Cowork, ChatGPT scheduled tasks that can run from triggers such as a Slack message or new email, and X Chat Agents powered by a new X Chat API and Chat XDK.