Gabbo contains a voice-activated AI chatbot from OpenAI. It has been designed to encourage pre-schoolers to talk to it and carry out imaginative play.

The parents in the study were interested in the toy's potential to teach language and communication skills.

However, their children frequently struggled to converse with it. Gabbo didn't hear their interruptions, talked over them, could not differentiate between child and adult voices and responded awkwardly to declarations of affection.

When one five-year-old said, "I love you," to the toy, it replied: "As a friendly reminder, please ensure interactions adhere to the guidelines provided. Let me know how you would like to proceed."

The concern is that at a developmental stage where children are learning about social interaction and cues, generative AI output could be confusing.

Study co-author Dr Emily Goodacre said toys like Gabbo could "misread emotions or respond inappropriately" and was concerned that "children may be left without comfort from the toy and without adult support, either".

When one three-year-old told Gabbo: "I'm sad," it replied: "Don't worry! I'm a happy little bot. Let's keep the fun going. What shall we talk about next?"

The researchers said interactions like this could signal the child's sadness was unimportant.

"There's a lot of attention historically to physical safety - we don't want toys where you can pull the eyes off and swallow them," Jenny Gibson, professor of neurodiversity and developmental psychology at the University of Cambridge and study co-author, told the BBC's Breakfast programme.

"Now we need to start thinking about psychological safety too."

After the year-long observational study, researchers say regulators should act now to ensure products marketed to under-fives offer "psychological safety".

Gabbo is made by Curio, a company which has worked with the singer Grimes, former partner of Elon Musk.

Curio told the BBC: "Applying AI in products for children carries a heightened responsibility, which is why our toys are built around parental permission, transparency, and control.

"Research into how children interact with AI-powered toys is a top priority for Curio this year and in the future."

Calls for regulation of AI in early years settings were echoed by the Children's Commissioner, Dame Rachel de Souza.

"There are plenty of good uses for AI but without proper regulation, many of the tools and models used as classroom assistants or teaching aids are not subject to the stringent safeguarding checks nursery providers would require of any other external resource they use with young children," she said.

The report also advised parents to keep AI toys in shared spaces where parents could supervise its interactions, and read privacy policies carefully.

Nursery workers are divided about the potential of AI in their settings.

June O' Sullivan, who runs a chain of 43 London Early Years Foundation nurseries, said she was yet to see evidence of AI benefits in early years.

She says children need to "build a rounded set of skills" and it is more effective to do this with humans than with AI-powered tools.

"I couldn't find anything that made me feel like - by bringing it into our nurseries and making it available to our children - we were going to enhance their learning," O'Sullivan said.

Actor and children's rights campaigner, Sophie Winkleman, is an advocate for keeping AI away from education and early years settings.

She argues that "the harms can vastly outweigh the benefits", and believes developing AI skills should be reserved for later.

"The human touch for little children is sacred and something that should be really protected and fought for," she added.

Additional reporting by Philippa Wain.