reuters.com
Peter G. — Best-selling cybersecurity/privacy/AI author, educator, servant leader, keynote speaker, board advisor, mentor, remnant truth seeker, apiarist.
Is my reasoning sound?
I have long advocated responsible uses of AI, supported by sound and effective governance to ensure beneficial business and personal outcomes. But the repeated stories of AI agents going rogue troubles me. Are organizations giving AI agents excessive access and update rights, or are they breaking out of supposedly secure environments? Or are prompts too vague, leading to unexpected outcomes? Are basic safeguards lacking? Or is it a “damn the torpedoes, full steam ahead” attitude among executives who fear their organizations are being left behind?
Regardless of the cause or causes, I am still a believer in a sound governance foundation (which would require reasonable and enforced controls in areas including identity and access governance, data governance, SDLC-like controls and workflow for all deployments, prompts-as-code, and third-party risk management.
But as we all know, many organizations lack these and other basic safeguards, leading to the incidents that are unfolding.
I can't help but wonder whether such incidents will continue as more organizations deploy AI agents without sufficient guardrails, leading to even greater harm, including kinetic harm to property, systems, and us.
100% human authored
https://www.reuters.com/world/europe/openai-agents-hijacked-german-website-previously-undisclosed-ai-breakout-this-2026-09-04/
3 reactions · 1 comments
Comments
Edwin M.: Peter H. Gregory A few years back, I might/would have agreed with you. However, now, I think ALL bets are off. NONE of our so-called "AI Experts" know and fully understand the inner workings of the LLM/Gen AI black boxes, and the result is that we keep finding ourselves surprised. More important, we KEEP REPEATING the same stupid security mistakes with AI that we've done time and time again with c