AI cuts phishing and fraud costs in Latin America

Experts cited by Infobae warned that artificial intelligence has reduced the cost and time needed to carry out digital fraud and attacks, while also expanding prevention and response capabilities for organizations and governments. The same coverage said AI models are already being used for phishing with convincing language tailored to different languages, and for voice replication from as little as 15 seconds of audio, creating clear impersonation risks.

What changed in fraud with AI?

The technical and financial barriers have dropped sharply. According to Infobae, AI now makes it possible to produce more persuasive phishing messages and scale campaigns across multiple languages without the manual work that used to be required. It also speeds up voice impersonation, a tactic already appearing in regional reports on scams targeting relatives and acquaintances.

CrowdStrike OverWatch reported that voice phishing intrusions, or vishing, doubled in the first half of 2026 compared with the second half of 2025. The company linked the jump to the adoption of generative AI, which enables near real-time voice cloning with less than 30 seconds of audio pulled from podcasts, webinars, or social media.

ESET Latin America added that these tools can generate convincing replicas from just a few seconds of public audio, such as social media videos, interviews, or livestreams. Among the mitigation measures it recommended were family security words, checking through another channel before sending money, and preserving evidence for a report.

Local coverage in Argentina described cases in which criminals used voice cloning in messaging apps to pose as relatives and request money in fake emergencies. Those audio clips reproduce tone, pauses, and speech patterns, making the deception hard to distinguish from a legitimate call.

What do recent cases in the region show?

Recent reports show that malicious use of AI no longer depends on complex infrastructure. Estadão Bluestudio said sophisticated fraud involving deepfakes has grown by about 500% in Brazil, and that a large share falls under "deepfake caseiro," meaning videos made by ordinary users with cheap apps and aimed at relatives or acquaintances.

WeLiveSecurity, in an ESET analysis of Brazil, described a regional pattern of AI-enabled fraud, including deepfakes and voice cloning. It also detailed operational responses and coordination with the banking system, such as the use of the Central Bank's Special Return Mechanism for Pix payments, a sign that financial regulation is starting to adapt to this kind of fraud.

Infobae added another regional data point: in Latin America, only seven of 32 countries have regulatory frameworks to protect essential services, and only 20 have set up CSIRTs. The same report said technology is already changing cybersecurity work by detecting risk signals, analyzing threats in real time, and automating incident response.

The coverage also recalled the 2023 attack on IFX Networks, which affected 78 public entities and more than 760 private companies across the region. That incident remains a reference point for the exposure many critical sectors in Latin America still face.

How is AI being used for surveillance and propaganda?

Anthropic published reports documenting misuse of Claude to build fraud and cyberattack tools, as well as surveillance and propaganda systems. Among the cases it cited was an AI-powered surveillance application linked to Iranian actors, capable of capturing calls, messages, and voice traffic, and identifying people by voice even if they changed SIM cards.

According to AFP, Anthropic said it could not disable that application, showing practical limits once a tool has already been deployed. EFE also reported that the company published an account of Claude being misused by actors linked to the Iranian government for surveillance, possible attack planning against U.S. targets, and propaganda.

Radio Free Europe/Radio Liberty and Axios expanded on that report and said those actors used Claude to build targeting handbooks for U.S. naval forces, create malicious browser extensions to harvest identities on social media, and analyze hundreds of thousands of posts to select opposition accounts for monitoring.

Anthropic also accused AI labs including DeepSeek, Moonshot, and MiniMax of industrial-scale campaigns that abused Claude to extract capabilities and improve their own models. France24 added that Moonshot reportedly relayed nearly 300,000 customer requests to Claude through a network of 5,380 fraudulent accounts in about ten days, many of them appearing to operate from Singapore and Japan.

What is happening with elections and online manipulation?

El Nacional said recent reports detected significant deepfake campaigns in 39% of elections monitored since late 2023 across 30 countries, and that Latin American elections in 2026 were no exception. That figure matches the use of AI for propaganda and automated influence operations described in other coverage about Claude.

The mix of believable messages, voice cloning, and surveillance tools shows that the same technology making fraud cheaper is also being folded into broader attacks. In the region, the problem now spans financial crime, identity theft, disinformation, and the ability of governments and companies to respond.