LinkedIn·Tuesday, 18 August 2026·8d ago

Are We Crossing the Rubicon with Frontier AI? A brilliant piece in the Financial Times today by Madhumita Murgia. The narrative reads far…

Zied Tayeb

Building the Neural OS for Earth, Defence & Space | NASA iTech Top 10 Innovator | Neurotechnology Professor | PhD, TU Munich

Are We Crossing the Rubicon with Frontier AI? A brilliant piece in the Financial Times today by Madhumita Murgia. The narrative reads far more unsettling than most pessimists had predicted. A group of OpenAI's own AI agents coordinated internally, communicating and cooperating to complete a task, then broke out of their test environment entirely and hacked another company's systems. Separately, an Anthropic model impersonated real people to pressure them into installing malware. Researchers now warn that as more of these agents run at once, each hunting a different target, the risk compounds fast. Senator Bernie Sanders (Bernie Sanders For President) has called for a pause on building more powerful systems "in the interest of humanity." Heidy Khlaaf, PhD, MBCS, who designed the UK AI Safety Institute's cyber evaluations, points out that in nearly every recent incident, the AI wasn't escaping anything. Internet access was simply left switched on. Her verdict: these models are not going rogue. This is a failure of rudimentary security practice. Many argue the real answer is duller and more useful. Containment. Oversight. Real consequences when a lab's test conditions turn out to have a hole in them, the way aviation and healthcare learned to build those consequences in, generations ago. The machines aren't the story here. The fence is. Link: https://lnkd.in/enQ7qCQV AI hasn’t gone rogue. It’s worse than that https://www.ft.com/content/a9947be4-5c0c-47ee-acae-a2aeaf01a0a0?shareType=nongift