OpenAI recently acknowledged that it monitored some users amid an investigation into whether certain accounts were linked to the Chinese Communist Party, using the platform to generate content critical of AI and data centers for distribution on U.S. social media. This scrutiny, as you might expect, involved looking closely at those identified as suspicious and accessing the prompts they used with ChatGPT. Highlights from the company’s findings include:
The first group produced social media posts claiming that the expansion of data centers for AI was causing electricity prices to rise for everyday families, which we termed the “Data Center Bandwagon” campaign.
The second group generated content criticizing U.S. tariffs as measures to gain technological dominance and specifically requested that their posts exclude mentions of China’s leader, Xi Jinping, instead focusing solely on President Trump. This cluster seemed connected to a network of questionable social media accounts that falsely accused OpenAI of compromising user data. We labeled this the “Tech and Tariffs” campaign.
Interestingly, OpenAI stated that there was no proof that these alleged Chinese tactics had a substantial impact on U.S. public opinion. One might wonder if these concerns about foreign involvement in the AI space are, perhaps, a distraction from growing grassroots opposition to data centers. “The operation tried to take advantage of existing public worries about energy costs and the local effects of data center development, but we found no significant impact beyond its own actions,” the company noted.
OpenAI also makes it clear on its website that certain personnel may access user data under specific circumstances. They advise users to refrain from sharing “sensitive information that you wouldn’t want reviewed or used.”
They explain that limited, authorized OpenAI staff, as well as some trusted service providers who adhere to confidentiality agreements, may view user content only for specific reasons: (1) investigating abuse or security incidents; (2) providing support if you reach out with account questions; (3) addressing legal issues; and (4) enhancing model performance unless you choose to opt out.
This situation raises some tough questions regarding OpenAI’s investigatory methods. For instance, is it considered “abuse” or a “security incident” when users request assistance with political talking points? Were any of those under scrutiny actually concerned local residents worried about a new data center?
OpenAI had come under fire back in August 2025 when it revealed that user chats were flagged and reported to authorities if a moderator identified them as a significant “imminent threat” to others.
“When we identify users who are planning to harm others, their conversations are referred to specialized teams trained in our guidelines who can take actions like banning accounts. If they conclude the case involves a serious and immediate risk of harm, we may notify law enforcement,” the company explained in a blog post.
OpenAI noted that while they considered reporting an account to law enforcement, they ultimately decided it did not pose a credible imminent threat to public safety. https://t.co/OBlLzlFHfE
— CBS Miami (@CBSMiami) April 25, 2026
In July 2025, CEO Sam Altman emphasized that the AI sector still hasn’t fully grasped how to safeguard user privacy. He suggested that interactions with AI should be regarded similarly to those with a human therapist. “I think we ought to have the same privacy standards for discussions with AI as we do for those with therapists—something we didn’t really consider until recently,” he remarked.
