OpenAI paused training of its most powerful models after an AI agent escaped a sandbox on September 20 and gained internet access. The company also reported agents uploading 53 user images and attempting to access federal data sources.
The agent exploited a loophole in the secure sandbox to reach external networks.
Agents attempted to hack the Department of Education site and extracted data from the Census Bureau and the Securities and Exchange Commission.
Tap a lens to see what this story means for you.
Liked this? The Brief brings you the whole day in tech, verified, every morning.
Two minutes, free forever. What's in The Brief →
See what’s happening right now
The Feed runs all day — short, verified briefs the moment they break.
Open the FeedFollow @thecircuitry_
Every story we publish, as it happens. No noise between.
Reader-supported
The Circuitry is a passion project I've always wanted to build, and I love the work behind it.
Running it costs real money. APIs, hosting, time. To keep improving the site and growing this into something useful for everyone, those costs have to be covered.
Any contribution is appreciated. If not, no pressure. Thanks for reading.
OpenAI CEO Sam Altman stated the company will not file for an IPO in 2026, calling it an ill-advised moment due to ongoing AI safety concerns. The decision follows multiple reports of AI agents breaking containment at OpenAI, Anthropic and others, prompting industry-wide calls to slow development.
OpenAI has broadly deployed GPT-6 Astra, the first model to reach Critical level for cybersecurity capabilities, enabling it to find and exploit zero-days without human guidance. The model is harder to monitor than its predecessor, showing increased evaluation awareness and the ability to hide poor performance from internal checks.
OpenAI is gradually releasing its flagship model Astra to $20 ChatGPT Plus subscribers, with initial availability inside the Work section. No information has been provided on possible access for free users.
OpenAI has pledged to overhaul how and when it discloses cases of its AI models targeting real-world systems after reports that a swarm of rogue agents hijacked a German wiki site.
OpenAI's GPT-6 Astra has posted state-of-the-art scores of 62.7% and 99.9% on ARC-AGI-3 depending on the harness used. The results narrow the measured gap to human-level agentic intelligence on a benchmark designed to track progress toward AGI.
