Users on platforms such as Baidu's Baijiahao shared photos of a person complaining to iFlyTek's customer service about how their child's device, a study assistant tablet, had described Mao as "narrow-minded" and "intolerant" for starting China's Cultural Revolution.

Reuters was unable to verify the account and iFlyTek did not respond to requests for comment.

A customer service representative on its website, asked about the incident, told Reuters that it had been dealt with.

Chinese media outlet Cailianshe cited iFlyTek's founder and Chairman Liu Qingfeng as saying that a supplier had provided the content for trial and that both the supplier and iFlyTek staff were punished after the problem was discovered.

The company had also strengthened the device's content review mechanism, he was also quoted as saying.

China censors material which may not be compliant with its policies, or is deemed potentially divisive or critical of its policies or leaders, even past ones. Mao, who died in 1976, is still officially venerated by the ruling Communist Party as the founder of modern China.

The iFlyTek incident highlights how generative AI can be unpredictable and can subvert the government's censorship tools.

China is going through a generative AI craze and the government has rolled out rules for firms looking to push products to the public. It has proposed the compilation of a blacklist of sources that cannot be used to train Chinese AI models.

IFlyTek on Tuesday revealed its latest Spark AI model which the company says can rival OpenAI's ChatGPT in most key areas.

Liu said iFlytek's model could compete with OpenAI's GPT-4 early next year.

The company also said that it is working with Huawei Technologies on an AI model training platform which uses Huawei's flagship AI chips.

(Reporting by Josh Ye; editing by Brenda Goh and Sharon Singleton)

By Josh Ye