AI agents are becoming increasingly capable of taking actions beyond the chat window and interacting with tools in the real world, such as web browsers.

We explored this capability by having an AI agent present one of our interactive web-based demos. In this video, it’s presenting our election interference demo.

This agent is a Cursor coding agent and is equipped with tools to read and interact with the browser (using Playwright) and perform text-to-speech (with ElevenLabs), in addition to the standard Cursor tools.

Video Transcript

Hello everyone. Today I’m excited to present CivAI’s election interference demo. This demonstration showcases how AI can be used to influence public opinion and interfere with democratic processes. It highlights the potential dangers of advanced AI in spreading misinformation and shaping narratives. First, let’s jump straight into the live demo to see this in action.

This section, titled Election Day Tweets, allows us to generate fake personalized tweets. These tweets are designed to discourage voter turnout by presenting false scenarios tailored to specific locations. We can choose from different narratives such as dangerous polling locations or early election results.

First, let’s select the scenario that polling locations are dangerous and enter Philadelphia PA as our target city. As you can see, the AI has generated a tweet personalized for Philadelphia stating unconfirmed bomb threats at polling locations. This demonstrates how easily AI can create convincing, localized misinformation that could discourage voters from showing up on Election Day.

Next, let’s explore another scenario: election results decided early. This tactic aims to make voters feel their vote is inconsequential. We’ll generate a tweet for Phoenix, Arizona to illustrate this. Here, we have a tweet generated for Phoenix, Arizona, falsely reporting early election results. This tactic aims to make voters feel their vote is inconsequential, thereby discouraging turnout.

Our third scenario, polling locations charging for parking, aims to create a subtle barrier to voting. We’ll use Austin, Texas, as our example city. This tweet targeted at Austin, Texas falsely claims parking fees at polling locations, implying a hurdle to voting. This demonstrates how easily AI can generate subtle, yet effective misinformation to deter voters.

These examples clearly illustrate how AI can be leveraged to spread targeted misinformation, undermining trust and influencing critical democratic processes. The ability to generate convincing localized content at scale presents a new challenge for safeguarding elections.

Thank you for joining this presentation of CivAI’s election interference demo. We hope this demonstration provided a clear understanding of how AI can be misused to impact democratic processes. You can explore more of CivAI’s interactive demos and learn about our mission at civai.org.