Artificial intelligence did not invent phishing, credential theft, or romance scams. What is changing is how easily criminals can adapt these methods, automate tasks, and try them at greater scale.
An Anthropic intelligence report published on September 10, 2026, describes cases in which operators used Claude for tasks ranging from vulnerability research and malware development to espionage and fraud. These are cases detected by the company, not a measure of how often this happens across the internet; this distinction is key to understanding what the report does and does not tell us.
In everyday scams, the most visible risk is that fake messages, voices, or images may seem more convincing. In more complex cyber operations, AI can also act as a copilot or coordinate tasks. In both cases, the technology lowers costs and saves time, but it does not eliminate the need for people, stolen data, or opportunities to attack.
More credible, personalized, and faster messages
Scammers can ask a generative tool to write or translate messages, adjust their tone, and prepare versions for different recipients. This can reduce language errors and make a fraudulent email or message look more like legitimate communication. The FBI also warns that generative AI is being used to facilitate fraud schemes and social engineering: manipulating people into revealing information or taking an action.
The same logic applies to impersonation. With audio samples or images available, criminals can try to imitate someone’s voice or appearance. A familiar voice urgently asking for money can pressure the victim to act before checking the story. But hearing a familiar voice or seeing a face on a video call does not, by itself, prove who is on the other end.
AI can also help manage many interactions at once: generating variations of a message, answering initial questions, or sorting potential victims. This does not mean every scam is fully automated. Often, a person still chooses the targets, monitors the conversation, or decides how to exploit the access obtained.
From assistant to operational automation
In a cyberattack, a model can help summarize public information, write code, analyze data, or prepare technical steps. If connected to tools that execute commands, search for exposed systems, or process files, it can become part of a more automated workflow. An AI agent is a system that, in addition to responding, can chain tasks together and interact with tools according to instructions.
Anthropic’s report describes operators who organized agents in parallel to investigate systems, search for vulnerabilities, test code in their own environments, and gather information. In some cases, tasks continued with little direct supervision. The company says it interrupted the identified activities and strengthened its safeguards; since this is a report from the provider itself, it should be read as a description of cases observed by Anthropic, not as an independent audit of the entire sector.
The most important change may be economic: if a tool speeds up research or analysis, a person can spend less time on each campaign or target more people. This does not automatically make every attack more sophisticated or more serious. AI can make mistakes, invent information, or generate faulty code; attackers still need access, expertise, and a vulnerable target.
What changes—and what doesn’t—in scams
The basic techniques remain familiar: tricking someone into clicking a link, reusing stolen passwords, exploiting unpatched software, or impersonating an institution. AI can make it easier to prepare an attack and tailor it to each victim, but it does not replace these entry points.
This also explains why some warnings may overstate the role of technology. A flawless text does not prove it was made with AI, and a scam can be effective without any generative model. Likewise, detecting an artificial image or voice is not always easy or reliable. It is better to check the request and the channel than to try to guess which tool is behind it.
How to reduce the risk from a suspicious message
- Verify through another channel. If someone you know urgently asks for money or codes, call them at a number you already have saved or confirm in person; do not use the contact details in the suspicious message.
- Pause when pressured. Urgency, secrecy, and requests for transfers, passwords, or one-time codes are reasons to verify before acting.
- Check links and permissions. Do not sign in through an unexpected link or grant account access because an unverified message asks you to.
- Protect your accounts. Use unique passwords, enable multifactor authentication when possible, and install security updates.
The most practical defense is not to trust in a foolproof detector of artificial content. It is to establish a pause and a verification method for sensitive requests. AI can make the initial contact more convincing; confirming who is making it and what they are asking for remains a human responsibility.
Sources and references
- Anthropic, “Detecting and countering misuse of AI: September 2026”
- FBI Internet Crime Complaint Center, “Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud”
- FBI Internet Crime Complaint Center, 2025 annual report
- FBI Internet Crime Complaint Center, alert about the Kali365 phishing kit
Image: TimeBusinessNews | CC BY-SA 4.0 | Wikimedia Commons.
Educational notice
This content is for educational and informational purposes. It does not constitute financial advice or an investment recommendation.