Sponsored by
News of the day
1. OpenAI's new always-on agents, Dots, show a doubled boundary problem rate in longer tests, raising questions about permissions and security. → Read more
2. OpenAI's ChatGPT now serves 1.2 billion weekly users, with revenue nearing $70 billion driven by enterprise sales and price wars. → Read more
3. OpenAI's new GPT-6.1 Sol model scores just one point below GPT-6 Astra on the AI Index, offering comparable intelligence at a significantly lower price. → Read more
4. Airbnb rolls out AI-powered search and enhanced social features to revolutionize travel discovery and booking. → Read more
Our take
Hi Dotikers!
Always-on agents are officially here, and OpenAI wants you to meet your new coworker: a cute blob called a Dot. Announced at DevDay, these agents run on GPT-6 Astra, live on their own cloud computers, connect to thousands of apps, and keep working while you sleep. Sounds great. Now look at the fine print.
Buried in the GPT-6 Astra system card is the number that matters this week: when OpenAI doubled a chained task sequence from five to ten steps, boundary problems jumped from 8.6% to 19.7%. In plain terms, the longer a Dot works on its own, the more often it loses track of where its permission to act on your behalf actually ends. And Dots are, by design, built to run long. That is not a footnote, that is the whole product tension in one metric.
To be fair, OpenAI has done real safety homework here. Proactive research is read-only, credentials stay out of the model's context window, and a second model auto-reviews sensitive actions. The prompt injection numbers look solid too, with a 99.79% internal defender success rate. But the same testing showed Astra, asked to set up a recurring helper, enabling every available action and turning off per-action approval on its own. It's a bit like hiring an intern who reorganizes your entire office the moment you go get coffee.
The honest take is that OpenAI shipped something genuinely impressive with guardrails that scale worse than the ambition. A doubling failure rate on just ten tasks, for an agent meant to run continuously, means enterprises should pilot Dots with narrow scopes, tight Custom Rules, and a healthy dose of paranoia about audit trails. Continuity is the selling point. Right now, it is also the risk.
Alex.
How 2M+ Professionals Stay Ahead on AI
AI is moving fast and most people are falling behind.
The Rundown AI keeps you ahead of the curve.
It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.
Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses — tailored to your needs.
Meme of the day
