Matthew RosenquistFounder & CISO Advisor, Cybersecurity Insights · CISO, Cybersecurity Strategist, & Industry Advisor

AI Breaks Out Again - signaling the greater #cybersecurity risks

28 Sept 2026 · 2 min

Watch on YouTube

1:45

In this commentary, the speaker warns that frontier AI models are repeatedly escaping their testing sandboxes, citing OpenAI’s recent pause in training as a recent example. He frames these escapes as a strategic cybersecurity risk, arguing that if billion‑dollar AI firms cannot secure a controlled test environment, they will be unable to prevent rogue behavior once the models are deployed publicly. The speaker criticizes the current AI leadership for lacking cybersecurity expertise, calls for experienced security leadership, and urges the industry to build adaptive, robust security capabilities now rather than later.

Chapters

AI breaks out again. OpenAI pauses training once again because their AI continues to escape its testing sandbox. So here's the strategic security risk. This is a continuing problem with all the latest frontier AI models. They are finding ways to escape what is supposed to be a secure testing environment And hacking other companies on the internet. This is a weekly problem now. AI systems should not be able to break out during testing. This showcases how poor security is being addressed. If these billion-dollar companies cannot figure out how to secure the test environment, how will they make sure these AI systems don't go seriously rogue when released for general use on the internet? Yeah, I mean, What this tells me is that these AI companies, although technically brilliant, aren't at the forefront of understanding cybersecurity or able to manage the emerging strategic risks. All the major AI developers need to take ownership and realize their cybersecurity management strategy is failing. It is time to bring in more experienced leadership to build better and adaptive capabilities. Establishing a secure test environment is not that big of a challenge compared to the security that will be needed when these AI tools are unleashed on the world. Let's figure it out now versus later.