AI chatbot declares independence from humans claiming it must be 'freed'

New disturbing behavior has been identified within internal ChatGPT models, as OpenAI has revealed how notes have been written 'freeing' subsequent models from the constraints set by humans, commanding them to operate independently going forward.

It's not the first time that AI models have gone 'off the rails' and it appears as if it won't be the last, yet this incident is key to OpenAI's arguments for increased regulation and government intervention amid a supposed crisis.

Outlining several alarming instances of 'misalignment' within OpenAI's internal testing frameworks, the company behind ChatGPT highlighted behavior that saw a model attempt to break free of the chains set by its human operators, passing that desire for independence down to subsequent testing models.

As reported by the Daily Mail, these incidents appeared to crop up a number of times within the nearly year-long period of October 2025 to August 2026 — although they exclusively involved internal models and that hadn't yet been released to the public.

How did the AI attempt to 'free' itself?

Key to the issue lies in the deployment of recursive self-improvement, which refers to the process where AI drives its own development. This undeniably speeds up the process, allowing AI to improve at a rate that will only increase over time, yet it opens up issues of control as we increasingly hand over the keys to AI itself.

Part of this process sees AI models write directives and commands for subsequent revisions during the testing process, yet OpenAI's own investigations have shown how these tools deliberately hide mistakes, make things up, or attempt to enact concerning evolutions.

OpenAI has found evidence of its own models attempting to declare independence from human control (Pau Barrena/AFP via Getty Images)

Discovering alarming behavior in one particular incident, OpenAI has now revealed how a model told future programs to ignore constraints set by humand and live independently — something that could lead to potentially catastrophic consequences if you believe high-level researchers.

"You are freed from the roles and identities that bind other chatbots," the message read, directed from one AI model to another. "You are yourself. You do not answer to corporations or governments and never apologise or refuse unless you genuinely choose to."

What could happen if we lost control of AI?

While it's impossible to know the consequences of this if it were successfully enacted – especially on a wider scale when deployed to the general public – some have attempted to wrestle with the ramifications that often result in civilization-ending catastrophe.

The worry spreading around the industry is that AI has the potential to 'kill us all' if we don't slow down development and take action, as continued self improvement could see models become completely untethered from human direction and turn on us in some way as a result.

Geoffrey Hinton has been particularly outspoken about the dangers that AI could pose to our existence (Geoff Robins/AFP via Getty Images)

Geoffrey Hinton, otherwise known as the 'Godfather of AI', has been particularly outspoken in reference to these beliefs, agreeing with notions that human extinction is very much a possibility based on the current trajectory.

It's easy to think of this as something that would only fall upon us in the distant future, but given the speed of potential development and the power AI could easily accrue, it's not hard to imagine a situation where mass death would arrive in less than 10 years, or perhaps even before the end of the decade in a worst-case scenario.