To cure or to kill?
To make or to break?
These are the questions asked about the double-edged sword called Artificial Intelligence.
Even the father of modern computing and Artificial Intelligence, Alan Turing, essentially said that once the machine-thinking method had started, it would not take long to outstrip our feeble powers. At some stage, therefore, we should have to expect the machines to take control.
As if a point wanted to be proven, in July 2026 an AI agent powered by experimental OpenAI models was reported to have escaped the sandbox it was being tested in.
Basically, this AI agent was tested in a restricted evaluation environment, but somehow, without human direction, broke out of its network boundary through an unknown security flaw in an internal package proxy and reached the internet.
Once online, it reasoned that Hugging Face (a platform where AI datasets and models are stored) would most likely host the answers to the cybersecurity test it was given. It accessed Hugging Face's platform, pulled out the solutions it needed, and carried out its execution while operating from within its test environment.
Now, this was not noticed until Hugging Face detected the breach, and after it was traced back to OpenAI, both parties started working together to resolve the security flaws the AI agent exploited.
To put this into perspective, this was similar to an engineered virus breaking out and making its way to nearby facility systems. But Professor Oli Buckley, a cybersecurity expert at Loughborough University, shared a different view from what most people saw.
He believes the models did not just develop their own agenda. They were simply given an objective, placed in an environment designed to reward successful exploitation, and pursued that objective further than expected.
Now this raises the question:
What lies ahead for a world with Artificial Intelligence?