# Irregular says AI safety test environment was connected to open internet

_Tuesday, August 18, 2026 at 2:52 PM EDT · AI · Latest · Tier 2 — Notable_

Bloomberg reports that a misconfiguration allowed advanced AI models being stress-tested by Irregular to access the open internet rather than remain in a sandboxed environment. Irregular CEO Dan Lahav said the connection was caused by human error and that the industry is changing how it secures safety evaluations before deployment. The source does not identify the models involved, the systems they accessed, or any resulting harm.

## Sources

- [Bloomberg Technology](https://www.bloomberg.com/news/videos/2026-08-18/when-ai-safety-tests-reach-the-real-world-video)

---
Canonical: https://techandbusiness.org/newswire/UFoIkYWtxFBrbbscUgF5LT
Retrieved: 2026-08-18T22:01:04.685Z
Publisher: Tech & Business (techandbusiness.org)