ChatGPT’s role went beyond writing articles; it also helped tailor them to the requirements of individual media outlets and draft messages persuading editors to publish them. Foreign influence campaigns no longer necessarily need to create fake news websites to spread propaganda. In this case, an operation originating in Iran successfully placed AI-generated articles in legitimate media outlets.
اضافة اعلان
According to an OpenAI report on a campaign dubbed “Bogus Bylines,” the company identified nearly 100 articles published or republished by around 12 media outlets worldwide. The articles focused on the conflict between the United States and Iran, although OpenAI was unable to determine precisely who was responsible for the operation.
AI turns influence campaigns into an integrated publishing operation
ChatGPT helped those behind the campaign prepare each article to make it appear consistent with the style and requirements of the targeted media outlet.
The operators asked the tool to review English-language drafts, compare them with the publishing requirements of different media organizations, and suggest necessary revisions. Once the revisions were approved, they asked ChatGPT to draft emails submitting the articles to editors and persuading them to publish them.
This approach made it easier to repeat the process across multiple media outlets.
The campaign used seven fake journalistic identities under whose names the articles were published. Some of these fictional personas were supported by social media accounts, while OpenAI noted that ChatGPT had written the biography of one of them.
Previous tests had also demonstrated how easily chatbots could be used to create convincing fake news stories.
Together, these elements made the submissions appear to editors like ordinary journalistic contributions.
The real vulnerability was trust in newsrooms
The campaign relied on the channels that media organizations already use to receive outside articles and contributions.
According to The Washington Post, one article reached the Port St. Joe Star through PeaceVoice, an organization that had maintained a working relationship with the newspaper for years.
The newspaper’s publisher explained that he had not independently verified the writer’s identity or the article’s source because of the trust established through that relationship.
This familiarity gave the article an initial degree of credibility, even before the actual party behind it had been verified.
Some media outlets later removed the published articles, while PeaceVoice tightened its policy on the use of artificial intelligence, according to The Washington Post.
OpenAI, for its part, banned the accounts involved in the operation as part of its ongoing efforts to disrupt malicious uses of ChatGPT services.
What should newsrooms verify now?
Relying solely on tools that detect AI-generated content will not be enough to address this problem. The vulnerability lies not only in the nature of the published material but also in the identity of the person supposedly behind the article.
Editors now have additional reasons to verify the identities of writers and contributors rather than treating professionally written biographies or recommendations from familiar organizations as sufficient evidence of credibility.
Related research has also shown that AI-powered chatbots can be manipulated into producing misleading information.
OpenAI said it had shared information about the campaign with the relevant authorities.
For editors and media organizations, the practical takeaway is to verify the identity of the person submitting a contribution and the channel through which it reached the newsroom, in addition to scrutinizing the content itself.
Resource: Al-Ghad.
