The use of artificial intelligence (AI) tools—such as ChatGPT and Gemini—is present in spaces that once seemed unlikely, such as the legislative processes of the Chamber of Deputies and the Senate. A recent investigation found that deputies and senators are increasingly using these technologies without clear regulation.

The story began when Sergio Bárcena, a research professor at the School of Humanities and Education (EHE) of Tec de Monterrey and founder of Buró Parlamentario, a civil organization that monitors and analyzes legislative work, collaborated with the team that will be in charge of drafting the Artificial Intelligence Law in the country.

That’s where the key question arose: are these systems already being used within Congress, and how often? To answer it, the first thing he and his team did was survey the advisors of the Mexico City Congress, who are responsible for a significant part of the technical work behind initiatives, positions, and speeches.

Following this, they discovered that its use was already common and growing rapidly, especially among younger advisors. They also identified that it was concentrated on specific tasks, such as summarizing texts and information. “I haven’t conducted these surveys at other congresses, but I suspect that usage is closely related to the capacity of the infrastructure,” says Bárcena.

Paradoxically, they discovered that the existence of AI hasn’t lightened their workload but rather increased it.

Afterward, they provided general training to Congress, where they found they shared concerns, such as authorship and the handling of sensitive data. “We all know what the problems are [with AI], but we’re still willing to compromise in exchange for it making our lives a little easier,” the researcher explains.

How they Detected the Use of ChatGPT in Legislative Speeches

This initial approach was limited by the fact that it depended on respondents accepting that they use these types of tools, so asking the legislators themselves didn’t sound like the best idea.

“Obviously they’re not going to say yes, because if you’ve seen it in the media, well, the legislator who uses ChatGPT gets lynched,” says Bárcena.

Faced with this difficulty, the team opted to look for evidence of AI in the parliamentary speeches themselves.

To do this, they developed a methodology to analyze speeches by members of parliament and senators to look for what they call “linguistic traces” of language models, that is, speech patterns typical of texts generated with these tools.

These footprints include:

• Mechanical enumerations: “first… second… third…”
• Recurring words: “clear,” “vision,” “strengthen”
• Linking formulas: “it is about…”
• Repeated structures: “not only… but…”

Using these indicators as a reference, the team reviewed speeches from 2021, before ChatGPT existed, and the result was consistent with their findings: from the second half of 2024, particularly from September onward, a significant increase in these traces was observed. This pattern suggests that the tool’s use in Congress not only exists but has grown rapidly in a short period.

“It covers both local and federal levels, so we already have indications to say, ‘Yes, you are using it; don’t lie to us; use it well,'” the expert emphasizes.

Risks of the Unregulated use of Language Models in the Legislative Process

“At first I thought this was neither good nor bad, until I realized there is a risk,” says Bárcena. For him, it is crucial that people who use it in legislation understand that LLMs (large-scale language models) were trained with data produced by people and institutions with cultural and ideological biases.

“If you don’t know that these models can generate different responses depending on the gender they perceive or the type of language they replicate, using them can carry a very serious risk,” says Bárcena.

For him, something as seemingly innocuous as a summary can reflect biases present in the data used to train the model, so its misuse can be dangerous. Furthermore, it’s important for citizens to know at what stages of the legislative process it’s being used, because otherwise, trust in representation is eroded.

Mexico Lacks AI Regulation and Guidelines for Congress

To date, Mexico does not have a general law on artificial intelligence nor clear agreements or guidelines for its use in legislative bodies, so taking that first step is essential.

“Let’s make agreements and declare its use so that the results are not governed by the algorithm but end up being governed by the human,” the researcher says.

The lack of clear rules contrasts sharply with the speed at which these tools are being integrated into legislative work. The study raises the need to define at what stages of the process they can be used and under what conditions.

“It’s impossible to stop using it; we’ve all seen the benefits it provides,” Bárcena concludes. “Let’s use it consciously and establish internal rules.”

Did you find this story interesting? Would you like to publish it? Contact our content editor to learn more: marianaleonm@tec.mx