OpenAI has identified and banned a set of ChatGPT accounts, likely originating from China, that were used to gather information on US persons, locations, and social-engineering methods. This activity was detailed in OpenAI's February 2026 report on disrupting malicious uses of AI.
The accounts prompted models in Chinese, were active during mainland Chinese business hours, and used VPNs to access the platform. They generated English email drafts for social engineering, purporting to be from a Hong Kong-based company named "Nimbus Hub Consulting." However, the use of Simplified Chinese characters in prompts, rather than Traditional Chinese, suggested the operators were based in mainland China. One account also generated an email from a Shanghai-based public relations organization. OpenAI named this operation "Silver Lining Playbook."
Key Points
- OpenAI banned a small set of ChatGPT accounts likely originating in China.
- The accounts requested information about US persons, forums, and federal building locations.
- Guidance on face-swapping software and generated social-engineering emails were also requested.
- Prompts were in Chinese, activity occurred during mainland Chinese business hours, and VPNs were used.
- Email drafts purported to be from "Nimbus Hub Consulting" and targeted state-level US officials or policy analysts.
- One account requested step-by-step installation guidance for FaceFusion face-manipulation software.
- The case study was originally published in OpenAI's February 2026 report.
Context
According to OpenAI, the generated email drafts resembled a social-engineering playbook for foreign intelligence services. These emails aimed to establish legitimacy for "Nimbus Hub Consulting," personalize and flatter targets by citing their public-sector background, stack incentives with vague but lucrative offers, and reassure recipients while attempting to move communications off-platform quickly to services like WhatsApp, Zoom, or Teams.
Why It Matters
This case highlights how AI models can be leveraged for information gathering and social engineering, posing risks to individuals and organizations, particularly in government and finance. It demonstrates the ongoing challenge for AI developers to identify and mitigate malicious uses of their platforms.
What To Do
- Note the tactics described for social engineering, such as establishing fake corporate identities and moving communications off-platform.
- Watch for similar patterns in unsolicited communications, especially those offering vague but lucrative opportunities.
- Review the February 2026 report from OpenAI for further details on malicious AI use cases and disruption efforts.
- Compare the described social-engineering techniques with warnings issued by democratic governments regarding foreign intelligence targeting.