We meet for lunch at Chatelaine du Liban, a Lebanese restaurant in Brussels’ Châtelain neighbourhood that has long earned its reputation. The room is bright, the welcome warm, and the lunch formula reliably generous: tabbouleh and houmous to start, kebab for me, shawarma for her, with muhammara, moutabal, and the restaurant’s much-loved orange blossom lemonade.

Across the table sits Caterina Rodelli, an EU policy analyst at Access Now. Often described as an AI policy expert, she resists the label. “People think I’m a technologist,” she says. “I’m actually a sociologist.” Her route into AI came via work on migration, border violence, and monitoring the growing use of data systems to decide people’s movements. “I didn’t come into this work to study technology,” she tells me. “I followed the harm. Technology was already there.”

That background shapes how Rodelli approaches AI: where others see novelty, she sees continuity.“AI is the latest layer in a much longer process,” she says. Long before AI became a headline issue in Brussels, she was examining how databases, biometric identifiers and automated border checks were being woven into Europe’s border regime. Migration, she argues, is not a marginal policy space but a testing ground. “It’s where technological systems are often deployed first,” she says, “tested on people with the least political protection.” 

Migrants, she notes, have long been subject to automated profiling, risk scoring and expansive surveillance infrastructures. AI, in her view, “has not so much introduced these practices as intensified them, accelerating political choices that had already been entrusted to technical systems”.

That perspective makes her sceptical of debates that treat AI as a rupture, detached from social context. “People talk about AI as if it came out of nowhere,” she says. What changed, she argues, was not the practice but its visibility. Seen this way, the EU’s flagship AI Act appears as a moment of exposure. Lauded as the world’s first comprehensive framework for AI, it promised to balance innovation with fundamental rights. For Rodelli, “the AI Act made visible systems that had been operating quietly for years”. Once negotiations began, she adds, “it became much harder to pretend these systems were hypothetical. They were already shaping decisions at Europe’s borders.”

Visibility, however, did not bring equal protection. As negotiations progressed, Rodelli says the AI Act revealed a familiar pattern: one set of rules for everyday life, another for the state. “What emerged very clearly,” she explains, “was the creation of two parallel legal frameworks.” Systems deemed unacceptable in most contexts were treated differently when deployed by law enforcement, border authorities or migration agencies. “There is one framework that applies to everyone else,” she says, “and another that applies to policing and borders, where exceptions multiply.” In the very areas where oversight is weakest and consequences most severe, the bar for restraint was set noticeably lower. Similar dynamics, she notes, have shaped earlier EU frameworks on data and security, from the GDPR to the EU’s interoperability databases.

In the case of AI, biometric surveillance offers a clear example. Practices such as remote biometric identification were repeatedly presented as red lines, with some uses restricted or banned in theory. But in practice, those limits came with caveats. “There was a lot of talk about bans,” Rodelli says, “but once you looked closely, law enforcement and migration authorities were treated differently.” Systems deemed too intrusive for everyday use remained permissible at borders and in policing -precisely the contexts where individuals have the least power to challenge them.

For Rodelli, these carve-outs reflect a long-standing tendency in how migration is governed at EU level. Border control, she notes, has often been treated as an exceptional policy space, where heightened security concerns justify broader discretion. “Migration is highly politicised,” she says, “and that creates a context in which exceptions are more easily accepted.” Safeguards applied elsewhere are more relaxed at borders, allowing certain technologies to be deployed under looser conditions. Over time, she adds, practices introduced under exceptional circumstances rarely stay there: “Once established,” she says, “they shape expectations, legal interpretations and technical standards, gradually influencing how similar systems are deployed in more ordinary settings.”

That recognition prompted a response from civil society. As the limits of the AI Act became clearer, Rodelli and others began organising beyond the legislative process itself. The result was Protect Not Surveil, a campaign bringing together digital rights groups, migrant rights organisations and racial justice advocates. The aim was to influence the AI Act, but also to keep attention on how surveillance technologies were being normalised across European policy. “AI was never the point,” Rodelli says. “It was the entry point.” 

For Rodelli, the debates around AI ultimately expose a deeper uncertainty about Europe’s priorities. “We keep talking about innovation,” she says, “but rarely about what we are actually protecting.” The repeated willingness to relax safeguards in the name of efficiency or security, she argues, reflects a broader political shift, one in which competitiveness increasingly outweighs human dignity and rights. In that sense, she sees the AI Act less as a solution than  as a mirror, revealing unresolved tensions at the heart of the European project itself.

Despite the structural challenges, Rodelli resists fatalism. What concerns her most, she says, is not even technology itself, but the shrinking space for imagination around it.“People are told this is inevitable,” she says. “That there is no alternative.” Working closely with migrant and racial justice groups has reinforced her belief that different choices are possible, if they are treated as political rather than technical questions.“The future is not written by algorithms,” she adds. “It is written by the decisions we decide to take responsibility for.”

The AI over Lunch interview series is a project part of Virtual Routes’ AI-Cyber Research and Policy Hub. If you would like to sponsor this series, please reach out to hu*@************es.org.

Have someone in mind we should interview? We’re happy to hear your suggestions!