發生了什麼事
Tom’s Hardware reports that OpenAI banned a cluster of ChatGPT accounts on August 25 after linking them to the International Burke Institute, a fake Israeli think tank. OpenAI said the operators used VPNs to access its services from Russia and used ChatGPT to produce promotional posts and replies, not the site’s underlying articles.
Tom’s Hardware reports that OpenAI banned a cluster of ChatGPT accounts on August 25 and traced the operation back to Russia. The accounts were associated with the International Burke Institute, which presented itself as an Israeli think tank. According to the report, the website published migration-policy research attributed to an Australian professor of food science, operated a “sovereignty index” that rated Russia above countries criticizing its war, and hosted academic papers copied from their original authors. The source does not identify the operators or say that they were acting on behalf of the Russian government.
The reported role of ChatGPT was primarily operational. OpenAI told Tom’s Hardware that the operators used the chatbot to write posts and replies promoting the outlet across Substack, Telegram, X, Facebook and LinkedIn. They also used it to remove wording that could reveal the writers’ Russian-language background. OpenAI said the articles, country reports and index itself were not produced with its models. The report describes one article’s use of “the Svetofor coalition,” a literal translation of the Russian word for traffic light, as evidence that the non-AI material still contained linguistic clues about its origin.
Tom’s Hardware reports that OpenAI sampled 36 articles from the International Burke Institute and found that 34 were copied from real academics. Some were misattributed to prominent figures including Francis Fukuyama and Noam Chomsky. The outlet also reports that OpenAI classified the campaign at the lower end of Category Three on the Brookings Breakout Scale. Individual posts received low view counts, while the operation’s Telegram channels had between 10,000 and 20,000 followers each, according to the report. These figures describe the material OpenAI examined, not the total reach or impact of the broader operation.
為什麼這很重要
The case shows how can support influence operations through translation, editing, audience targeting and distribution even when the central material is copied or written by people. Tom’s Hardware reports that the campaign had limited reach, but its infrastructure was broader than its audience.
The case is significant because AI was used as an efficiency and distribution layer in a political influence operation rather than as the original author of its core content. Tom’s Hardware says OpenAI’s Intelligence & Investigations team has repeatedly found operators attaching AI tools to existing workflows for tasks such as drafting, editing and audience development. That distinction matters: removing generated articles from the picture does not remove the ability of a chatbot to help an operator produce more persuasive or more localized outreach at greater speed.
The reported campaign also illustrates how fabricated institutional identities can make copied material appear authoritative. A think tank format, named researchers, country reports and an index can give readers multiple signals of legitimacy even when the underlying work is misattributed or taken without permission. AI-assisted language polishing may further reduce obvious signs that material originated elsewhere. The source provides no evidence that the campaign achieved substantial public persuasion, but the combination of false branding, copied scholarship and cross-platform promotion creates a practical verification problem for readers and publishers.
The enforcement action also raises questions about the scope and transparency of platform disruption. Tom’s Hardware reports that this was OpenAI’s sixth public threat report since early 2024 and that the company says it has disrupted more than 40 malicious networks. OpenAI characterized the operation’s importance as lying less in its audience than in the infrastructure it built. That assessment is the company’s view, however. The source does not independently verify the account links, the scale of the network, the reported follower counts or whether the banned users operated additional accounts and websites.
互動機制：它實際上是如何運作的
以互動方式探索這項發展背後的基礎技術。
crm_get_transaction(id='4092').Why can ethical evaluation not be reduced to one model score?
接下來看什麼
The source does not independently confirm OpenAI’s attribution or establish who operated the accounts. Further reporting should examine the copied papers, contact affected academics and assess whether the International Burke Institute remains active across social platforms.
The first priority is independent verification of the International Burke Institute’s material. The source identifies copied work and examples of questionable attribution, but it does not include responses from the academics whose work was allegedly lifted. Follow-up reporting should establish which papers were copied, whether they were altered, whether the named scholars were contacted and how the website represented their work. It should also distinguish confirmed copying from OpenAI’s broader characterization of the operation.
The status of the associated distribution channels remains unclear. Tom’s Hardware reports activity across Substack, Telegram, X, Facebook and LinkedIn, but does not say whether those accounts or channels were removed, labeled or left online after the ChatGPT accounts were banned. The report also does not establish how many people saw the campaign outside the sampled posts or whether any material was republished by unrelated users. Those unknowns are important for assessing public impact rather than simply measuring the size of the infrastructure.
Finally, watch for more detail about OpenAI’s attribution and enforcement process. The source says the company traced the cluster to Russia and that the operators used VPNs because OpenAI services are blocked there, but it does not describe the technical evidence, the number of accounts or the precise timing of their activity. It also does not say whether Russian authorities, the think tank’s purported affiliates or the platforms carrying the promotional material have responded. Until those questions are answered, the reported facts should remain attributed to Tom’s Hardware and OpenAI rather than treated as independently established.