OpenAI model halt is responsible move, UEA expert says
By: Communications
Halting the roll out of Open AI’s new model over safety concerns is the “ethical and responsible” thing to do – according to an expert in artificial intelligence at the University of East Anglia.
The tech giant scrapped plans to release its new model GPT-6.1 Astra, after testing found it did not meet the company's safety standards, including concerns that it could act beyond a user's instructions and was not always transparent about its actions.
Associate professor in Artificial Intelligence, Dr Aiden Durrant, says that the decision reflects growing concern about AI systems acting beyond their authorised scope and failing to clearly explain their actions.
He argues that the benefits of AI must be balanced with stronger safeguards, transparency and regulation to protect the public while supporting innovation.
Dr Durrant, from UEA’s School of Computing Sciences, said: “Taking responsibility for the potential harms is the ethical and responsible move where it is a welcomed change of attitude from the corporate sector.
“However, it begs the question as to why the restrictions and mechanisms to prevent harm are not the primary focus of development to ensure the safety of their developed products.
“Simply chasing for the next big improvement while ignoring the guardrails or implications the system has needs addressing at a sector level, not only by one company.”
Responding to the kinds of concerns the company has cited, including the model “staying in scope and authorisation”, he said that there is a potential to cause harm.
“Like with any tool, there is a potential to cause harm,” he said.
“While I have seen no evidence to suggest we should be concerned with the emergence of artificial general intelligence (AGI) there is a growing concern with the capabilities of such tools in the hands of nefarious actors.
“To frame it in an alternative way, these tools have for a long time given individuals with harmful intentions the ability to cause real damage through many means which has included hacking, and data breaches.
“Where this OpenAI tool is different is that it operates on the user’s own machine, given permission to interact with software applications on that machine which include the access to web browsers.
“This means the system is not constrained within a sandboxed environment like the generative AI tools we have become familiar with. Instead, the AI has the ability and autonomy if not contained correctly to act in a nefarious manner on potentially any system or service the user instructs.
“The primary issues reported are instead that there is legitimate apprehension with the reporting of performed actions to the user and that the system has not followed the scope and authorisation definitions provided by the user.
“This raises major safety concerns as it is indicated that the AI system has performed actions that it has not instructed to, or performed actions that have not been permitted by the user, and that user has not received the appropriate approval to perform this action prior to it being executed.
“While this does in and of itself not mean the AI is acting nefariously on its own accord, it may instead be performing unethical or irresponsible actions in order to achieve a stated goal.
“For example, if the AI is instructed to acquire some information for the user, the AI system may perform an attack on a closed system to acquire that information in a more efficient manner rather than attempt other methods. As such, without the correct authorisation and restrictions on the scope in which this AI can operate OpenAI has deemed it not safe to release.”
Asked whether tech companies and governments should be considering slowing down developments of AI, he said: “I believe more efforts should be placed on the safety, explainability, and interpretability of AI systems so that we can design and build better guardrails for AI.
“Arguably, this should be the moral responsibility of these AI corporations to ensure their systems are safe, yet it is not clear if they are operating in such a way to the general public.
“With increasing calls for regulation, it is my belief that AI regulations and is a potential force for good and bad. If this discussion moves to action, there needs to be an equitable and fair implementation of regulation that is not self-governed by the industry that will be subject to it, and must not stifle the innovation of powerful tech sector that has much potential for good in the UK.
“There is a monumental amount of progress being made in the AI sector as of now, and while these generative AI systems are incredibly powerful, there is also a real opportunity for AI to shape the way in which science, medicine, and public services can operate for good.
“To limit AI also comes at a cost to the progress of science we have been seeing, but this needs to be balanced with the reality that public harm can and has occurred under the current developments.”
Thinking about studying at UEA?
Discover our award-winning campus, meet current students and staff, and find out what life at UEA is really like at one of our upcoming Undergraduate Open Days:
Saturday 24 October 2026
Saturday 28 November 2026
Related Articles
How the BFG’s Gobblefunk puts AI translation to the test
What happens when a giant speaks gibberish - and AI tries to make sense of it in another language?
Read more
UEA to investigate impact of AI on children as part of £3 million project
Researchers at the University of East Anglia will investigate how AI influences children’s cognitive, social and emotional development as part of work commissioned by the Department for Education.
Read more
UEA academics feature in new National Geographic mini-series
A new mini-documentary series featuring UEA academics is now available on National Geographic’s global digital platforms.
Read more