European AI tool intercepts scam calls before the phone starts ringing

A small language model running on a user's phone serves as a protective layer between fraudsters and recipients.

Rita Barbosa, a founder of the Lisbon-based startup Guardian Labs, saw firsthand what falling victim to a phone scam can do to a person.

Fraudsters convinced her grandmother that the water at her home wasn’t safe to drink, selling her extremely overpriced water filters and resulting in a loss of over €4,000 (over $4,600).

Besides the money, what really impacted us was her losing her independence. She stopped picking up her phone. She became scared of using her phone and always asked me to make sure that she was doing everything correctly,

Rita Barbosa, founder of Guardian Labs

After Barbosa learned more about phone scams, she realized the problem wasn’t limited to Portugal or to elderly individuals.

This inspired her to develop the AI-based tool Guardião, which blocks potential calls before they reach users’ phones.

A 2026 report by the Global Anti-Scam Alliance states that 3 in 4 European adults have encountered a scam in the past 12 months, with phone calls and text messages among the primary entry points.

Globally, $442 billion was lost to financial fraud in 2025, an amount exceeding California’s state budget.

How does Guardião work?

Guardião is a very small AI model that runs on a user’s phone and scans incoming calls and messages in under 100 milliseconds, looking for scam clues like urgency and social engineering techniques.

Each message and call is fingerprinted to extract the attacker's artifacts, such as the sender, URLs, and payment identifiers.

Messages and calls identified as potentially fraudulent are intercepted before they appear on the phone.

Despite rapidly improving cybersecurity tools, phone scams remain difficult to tackle. There is also a challenge in developing algorithms to reduce false positives in systems designed to detect them.

For Barbosa, the real danger is false negatives – the messages and calls that pass through the system’s filters.

Nevertheless, Guardião has a system in place to prevent users from missing important messages or calls. Each time a potential scam is intercepted, the user is immediately notified, with an explanation of why it was blocked.

“The links are not clickable inside the app, so you don’t make the mistake of immediately opening them,” she tells Cybernews.

Training data comes from the police

In Portugal, where Guardião will soon be released, the model was trained on large volumes of fraud reports from the country’s Public Security Police.

Barbosa is already eyeing an international rollout of her scam-protection tool and is in talks with law enforcement in other countries.

However, the release outside Portugal would require additional training tailored to scam techniques specific to each country.

Even in countries that have the same language, such as Portugal and Brazil, scams can be completely different,

Rita Barbosa, founder of Guardian Labs

The Guardian Labs, which partners with 2 Portuguese universities and is publicly endorsed by the president's office, is also developing real-time intelligence to stop fraud at the corporate and institutional levels.

Barbosa says the platform will provide banks, telecom operators, and law enforcement authorities with intelligence on active international campaigns, enabling them to be blocked before any damage is done.

Fighting AI-powered scams with AI

The explosive use of generative AI following the release of ChatGPT in 2022 has opened new possibilities for fraudsters.

The technology is now used to generate realistic voice clones, scale up personalized messaging, and fake documentation, among other things.

AI development is also accompanied by growing concerns about cybersecurity, privacy, and the environment.

Barbosa says she understands the negative reputation AI as a technology has earned, but emphasizes that Guardião, as a small model, doesn’t use any data centers, because it runs directly on the user's phone.

“Nor is it giving back information to a server,” she says.

The app will be available free of charge to ensure it is accessible to everyone, regardless of their ability to pay.

Barbosa tells Cybernews, “We really believe that this type of protection is a human right.”