OpenAI Halts AI Propaganda Drive In The Philippines
AI powered Operation High Five generated pro government and anti Vice President Sara Duterte comments using ChatGPT for Facebook and TikTok. Inquirer file photo
OpenAI, the US company behind the popular AI chatbot ChatGPT, has banned accounts in the Philippines. These accounts were found to be using ChatGPT to generate a large number of social media comments. The comments generally supported President Ferdinand Marcos Jr. and criticized Vice President Sara Duterte, his former running mate.
OpenAI Takes Action Against Misuse
In a report titled "Disrupting malicious uses of AI" released on June 5, OpenAI detailed its actions. "We banned ChatGPT accounts that used our models to generate bulk volumes of short comments in English and Taglish," the report stated. These comments were posted on TikTok and Facebook and focused on political issues and current events in the Philippines.
The report identified Comm&Sense Inc., a public relations and marketing firm based in Makati, as one of the entities involved in this AI-driven influence campaign. As of this report, Comm&Sense Inc. had not responded to requests for comment.
You can learn more about the broader concerns of AI misuse in related discussions.
Uncovering Operation High Five
OpenAI dubbed the campaign "Operation High Five." This name was chosen due to the frequent use of emojis in the social media comments and their generally positive tone.
"The comments this operation generated and posted online were brief but partisan," OpenAI's report noted. "Typically, they praised President Marcos and his initiatives, or criticized Vice President Sara Duterte."
Despite its activities, OpenAI assessed the influence operation (IO) as having a low impact, classifying it as a Category 2 activity on their IO impact scale, where Category 1 is the lowest and Category 6 is the highest.
The Mechanics of the Influence Campaign
The report explained that those behind the operation used ChatGPT in several ways to support their political influence campaign targeting Filipino audiences.
First, ChatGPT was used to analyze social media posts about political events in the Philippines, particularly those involving President Marcos. It also suggested themes for replies to these posts.
Second, the AI model was instructed to generate large quantities of short comments, usually around 10 words long, based on the proposed themes.
Third, ChatGPT was employed to create public relations pitches and statistical analyses for the operation, potentially for existing or future clients. "These pitches stated that the operation had created five TikTok channels aimed at promoting President Marcos’ agenda," the report revealed.
OpenAI's investigation found that these five unnamed TikTok channels started posting videos in mid-February. This timing coincided with the beginning of the campaign period for the 2025 midterm elections. The AI-generated comments were then posted in response to these videos by numerous TikTok accounts. These commenting accounts typically followed no other accounts and had very few followers, ranging from zero to ten. "This commenting activity may have been designed to make the TikTok channels look more popular than they actually were," OpenAI suggested.
The same pattern of comments appeared on Facebook, replying to news reports from mainstream media outlets. OpenAI discovered that the Facebook accounts sharing these comments usually had zero to fifteen friends. These accounts were likely created around mid-December 2024, based on their initial profile picture uploads.
However, the engagement with these comments was minimal. "None of the comments that we identified online during our investigation received more than single-digit replies, likes or shares, and most received none at all," OpenAI stated.
Limited Impact and Platform Policies
OpenAI concluded that the goal of mass commenting on Facebook to flood comment sections was only "partially successful." For example, while the operation generated several hundred comments on a single Facebook post by ABS-CBN News, that post itself garnered around 23,000 comments in total.
On TikTok, where comments were focused on the five specific channels, video view counts varied significantly, from over a million views to fewer than 100.
Both Facebook and TikTok have policies prohibiting inauthentic behavior intended to artificially inflate engagement on their platforms.
