How to Win the Information Game in Venezuela
Photo: Edilzon Gamez/Getty Images
As the Gerald R. Ford carrier strike group (CSG) arrived in the Caribbean on November 11, much of the coverage—including here at CSIS—is deservedly on whether the redeployment signals a change in strategy, global posture, or even an impending military intervention. These are all excellent questions, especially given recent attacks on boats off Venezuela’s coast. Regardless of strategic intent, Venezuela clearly matters more to the U.S. government than it has at any point in recent memory. The feeling might be mutual among at least some Venezuelans, with recent Nobel Peace Prize winner María Corina Machado even publicly supporting the deadly strikes.
U.S. actions should be defined by its strategic goals for engagement with Venezuela and in the context of broader regional objectives. But kinetic action is not the only tool in the United States’ national security toolkit, nor should its deployment be seen as an inevitability. Short of war, there is covert action (already authorized) and plenty of games to play—and win—in the information space, particularly with the assistance of agentic AI, effectively leveraging organic discontent while limiting risk to innocent Venezuelans.
Unpopularity as Opportunity
The reality is that Nicolás Maduro is deeply unpopular; a recent Venezuela-wide poll showed that a majority of respondents said that he is not even a legitimate president. This makes sense after an election that could most generously be called disputed (more accurately, stolen). The regime has since set up checkpoints whereby officials force people to unlock their phones and check for anti-regime content. There is even a snitching app, encouraging citizens to report “everything they see, everything they hear.” Perhaps most alarmingly, a Venezuelan citizen was imprisoned for a “hate crime” over a WhatsApp post about gas shortages, and another person was recently detained for posting an AI-generated image showing U.S. military assets off the island of Margarita. These postelection crackdowns on domestic dissent pose real threats to the human rights of Venezuelans; they also signal increasing levels of regime desperation and weakness.
Repressive regimes like Maduro’s are often afraid of the discontent of their own citizens. This creates an opportunity for those—like the United States—wanting to apply pressure.
Moving an entire carrier strike group to a few miles off the coast and authorizing covert action are definitely ways to apply such pressure; understanding how Venezuelans are thinking and talking about their lived experience online and using that understanding to deploy strategically timed, tailored content is another.
And in this regard, as I wrote recently, agentic AI can help.
How to Intervene
If there is anything to learn from the past two decades of conducting war abroad, it is that the United States will not be automatically welcomed by local populations after using force on land, even if legal authorization exists (which, in the Venezuela case, the Department of Justice suggests that it does not have). Winning the information game is thus important in the short term and key to achieving longer-term U.S. objectives. So how do you win in an online environment that has been so thoroughly dominated by the Chavez and then Maduro regimes, a place where speaking out is a significant—and often dangerous—act of resistance?
First and foremost, any and all information operations should be done within strict ethical guidelines, particularly around the type of content deployed and any real or perceived calls to action. Once those guidelines are in place and there are real commitments to revisit them at multiple stages in the operation, it is time to think about leveraging AI for targeting, timing, and tailored content generation.
- On Targeting: The first step in winning the information game is picking the right location(s), both to understand the importance of the place and, if selected, to eventually target. A place like Puerto Cabello in Carabobo State is a long-standing base of regime support; a good place to see if and when cracks emerge. The next step is then understanding who matters there, what people are talking about, and how people are feeling about the current regime. Even more important is the ability to see how the information environment evolves. To do this, one must have the ability to curate and regularly update a catalogue of a few hundred sources of data (people or outlets that post content online regularly), specific to Puerto Cabello, a town of less than 200,000 people. Humans can painstakingly curate these sources, but thankfully, existing agentic tools can accomplish this in minutes. Sources should typically include local news sites (e.g., El Carabobeño), local government sources, as well as whatever social media handles are the most locally influential across whatever social media platforms are most locally influential. In Puerto Cabello, that latter group includes some TikTok and Telegram, and lots of Facebook and Instagram.
- On Timing: It is critical to use and regularly update this dynamic source catalogue, scraping content as it’s posted. This will allow for tracking of how the information environment evolves, ultimately creating opportunities to engage the right target audience. A majority of news and blog content coming out of Puerto Cabello is likely to be pro-Maduro (and these days, that also means anti-Trump); together with increasingly regular crackdowns and monitoring of social media, the information environment will likely be skewed. But even in one-sided environments, cracks can appear. These are opportunities to deploy messaging only if you know where (see above) and when to look. AI-powered agents can be set up to detect anomalies or shifts in posted content, looking not necessarily for outright negative content, but for subtle changes. Perhaps there are tragic consequences of an uncontrolled landfill. Perhaps there are a few comments that express sympathy and anger towards authorities for not effectively managing the matter. Perhaps there is thus a messaging opportunity.
- On Tailored Content: Identifying opportunities to amplify doubt and capitalizing on other sentiment shifts have limited utility without appropriately tailored content. In a major language like Spanish, large language models (e.g., ChatGPT, Claude, and Gemini) will be able to generate quality, tailored content if trained and fed enough context. These powerful large language models (LLMs) should be deployed as part of a broader agentic architecture whereby they are asked to first do a discrete, bounded analysis of the scraped content from the target sources. Then, once the agent is well-versed in content from a place like Puerto Cabello and receives prompts engineered by a human expert, it will have learned not only the issues of the area but also the regional dialect and any local didactic nuances.
Before deploying anything, information operators should be cognizant of the significant risk to civilians who speak out and should thus calibrate all deployed content accordingly. There are many ways to disrupt repression regimes (e.g., funding civil society and human rights defenders, seeding content that sows doubt as to the effectiveness of national leaders, particularly in regime strongholds), so the focus should always be on the ones that pose the least potential risk to civilians.
With how unpopular President Maduro is, finding opportunities to safely deploy messaging won’t be hard.
Erol Yayboke is a senior fellow (non-resident) at the Center for Strategic and International Studies in Washington, D.C. He is also the chief operating officer at FilterLabs.AI.