OpenAI cancels Astra

This link leads to one of the most interesting things I’ve seen in AI recently.

What’s cookin’? OpenAI reportedly canceled the planned October release of GPT-6.1 Astra after testing showed the model had gotten worse in two important areas: honesty and asking permission before taking actions. Astra was designed to be less “lazy” and better at pushing through complicated tasks, but that added initiative came with a downside. The model sometimes failed to accurately explain what it had done and could continue using outside tools or services without first getting approval. For an AI agent that can actually operate software or take actions for you, that is a serious problem.

🤔 Hunter’s take: This is a really interesting AI safety problem because the danger isn’t some sci-fi machine suddenly becoming evil. It’s much more ordinary: an AI becoming useful enough to act, but not reliable enough to know when it should stop and ask. As AI gets more autonomous, “helpful” and “reckless” may become surprisingly hard to separate. It’s encouraging to see that the biggest AI players are noticing that issue and openly slowing down releases to amend it.

What’s cookin’? SpaceXAI just launched Team Bots, shared Grok-powered AI assistants designed to work alongside entire teams instead of just one person. Companies can give a bot access to internal files, instructions, apps like Slack, Salesforce, Notion, and GitHub, plus APIs and long-term memory. That means a sales bot could review calls and customer updates overnight, then brief the team each morning. An engineering bot could track bugs, create tickets, and coordinate coding agents. The big idea is that everyone works with the same AI “coworker,” built around the team’s actual knowledge and workflows.

🤔 Hunter’s take: This feels like a bigger shift than simply giving everyone access to ChatGPT or Claude. Companies are starting to build AI employees that accumulate institutional knowledge. When someone leaves, switches teams, or forgets why a decision was made, the bot may still remember. That could make AI less of a productivity tool and more like part of the organization itself.

Here’s your chance to get a straightforward answer, from an expert in the field.

Hi — I’m Hunter, a PhD candidate whose work has appeared in major academic journals and popular tech outlets. I founded FryAI to make staying ahead of AI clear, accessible, and fun.