A new artificial intelligence tool, described by its own engineers as “too powerful for public release,” has been quietly deployed on the open web, prompting a formal investigation by the UK’s AI Safety Institute. The model, codenamed ‘Orion-X,’ was developed by a US-based startup and reportedly can generate realistic video, code, and text with minimal oversight. The company claims it was intended for “limited beta testing,” but security researchers soon discovered the system was accessible to anyone with a standard web browser.
Critics argue the release is a dangerous end-run around responsible AI governance, while the startup insists it was a “configuration error” that has now been resolved. The UK regulator is now examining whether the tool violates voluntary codes of conduct agreed at last year’s Bletchley Park summit. This incident underscores the growing tension between rapid innovation and safety protocols, and raises urgent questions about the public’s digital sovereignty in an age of accelerating AI capabilities.
