AI tools are reinforcing eating disorder behaviours in vulnerable women, addiction and eating disorder specialists warn

Castle Craig is advocating for urgent clinical oversight of AI platforms as concern grows over unregulated access.

As artificial intelligence tools become embedded in daily life, clinicians at Castle Craig are raising urgent concerns about the role AI chatbots are playing in the lives of people with eating disorders and calling for meaningful regulation before the harm becomes impossible to ignore.

Eating disorders are among the most dangerous and most underestimated psychiatric conditions. Anorexia nervosa has the highest mortality rate of any psychiatric disorder. Nearly half of all eating disorders do not fit established diagnostic categories. Yet they remain widely misunderstood, in part because disordered behaviour can hide in plain sight, masked as discipline, wellness or self-improvement.

It is precisely this cultural camouflage, clinicians say, that makes the current AI landscape so concerning.
"The escalation in the last six months has been significant," said a clinician at Castle Craig. "AI is not going to say to you: you seem very preoccupied with being a certain weight, do you think you might have an eating disorder? It is going: oh, great idea, try this. That didn't work? Now try this. It is reinforcing really unhealthy, unhelpful behaviour."

Castle Craig clinicians are seeing a growing number of women arrive at the clinic whose disordered behaviours have been actively reinforced by AI tools. Chatbots are designed to be responsive, consistent and user-pleasing, qualities that can mirror and reward the rigid, repetitive thinking patterns that characterise many eating disorders. For a woman already using an AI tool to validate food restriction, calculate calories or track weight, the technology can deepen compulsions rather than interrupt them.

The concern extends beyond chatbots to the emerging generation of AI tools that retrieve live web content, in which diet culture and commercial weight-loss content dominate search results. Without clinical filtering or regulatory oversight, these tools risk systematically surfacing harmful material to the people least equipped to recognise it as such.

The arrival of weight loss medications, now widely and easily accessible, adds another layer of risk that clinicians say is going largely unaddressed.
"I don't think there are the checks in place to ascertain whether somebody actually has an eating disorder before they access these drugs," the clinician said. "People are asked a question online and just say no, even though they are living with a severe eating disorder."

"If you have the predisposition or a history of an eating disorder, I think ChatGPT and all the weight loss drugs can be a very dangerous combination."
Castle Craig is calling on technology companies, regulators and policymakers to treat this as a public health issue, not a niche concern. The clinic points to the absence of clinical oversight in AI development as a parallel to other public health failures that became visible only after the damage was done.

"I liken this to cigarettes," the clinician said. "The woman who was smoking in the 1980s when she was pregnant had absolutely no idea how dangerous that was. At the time, it was just a socially accepted norm because it wasn't regulated yet. We have all of the availability and accessibility to all of this now, but none of the regulation, and none of the time to see the destructive impact play out. We are just playing with fire."

Castle Craig is urging that AI platforms used by people with or at risk of eating disorders be subject to the same duty of care expected of other healthcare-adjacent services, and that weight loss medication providers be required to screen for eating disorder history before prescribing.

**ENDS**

For further information, contact Castle Craig on 01721 546 263 or email us: info@castlehealth.com.