Artificial intelligence, an increasingly indispensable tool for those who work, could be used not only by employees, accounting and creative, but also by IT criminals: a decidedly particular “workers” category. The US company Anthropicwhich develops the Claude linguistic model, has recently published a report in which it claims that its technology has been exploited by attackers for cybercrime activities at a high level of complexity. The cases identified range from the writing of code harmful to the organization of real extortion campaigns, up to fraud orchestrated by groups related to North Korea. The company has reported how these episodes represent a qualitative leap in the techniques used by criminals, because AI allows you to automate decisions, reduce attack times and make intrusions more difficult to detect. In other words, AI is the new weapon used to perform cybercrosicswhich makes us all a little more vulnerable, at least potentially.
How cyber criminals used the AI for their purposes
The question is not limited to the only theft of data or ransomware, the malicious software that block systems up to the payment of a redemption: artificial intelligence is also useful for manipulating people with targeted social engineering strategies, as in the so -called “Vibe Hacking”that is, the use of AI to understand how to psychologically persuade a victim. Regarding this technique, Anthropic researchers explained that they had found a computer criminal who used Claude Code to commit large scale thefts and personal data. The author of the attack has affected at least 17 different organizations, including healthcare facilities, emergency services, government institutions and religious organizations, threatening to make public data to try to extort to the victims the payment of ransoms that in some cases exceeded half a million dollars in Bitcoin. According to Anthropic, the mind behind this attack used the AI following unprecedented dynamics. The company explained:
Claude Code was used to automate the reconnaissance, collect the credentials of the victims and penetrate the networks. Claude was authorized to make both tactical and strategic decisions, how to decide which data to subtract and how to formulate requests for extortion targeted from a psychological point of view. Claude analyzed the stolen financial data to determine the appropriate amounts of the ransoms and generated visually alarming redemption requests that were displayed on the victims of the victims.
The AI, therefore, did not work as a simple support tool by the IT criminals that used it to perpetrate similar attacks but, in fact, it had a strategic support role: it suggested which data to steal, how to exfilter them, and even how to determine the most convincing amounts to be requested as a redemption.
Another documented episode concerns the creation of fake profiles to access working positions remotely at large US technological companies. According to the report, North Korean operators used Claude to write applications, translate texts and even develop code. This allowed them to overcome cultural and linguistic obstacles that in the past limited their range of action. Once taken, these IT criminals could access the internal systems of companies, often without arousing immediate suspicions. Anthropic considers this method a new phase in scams related to the world of employment, because the use of AI drastically lowers the entrance barrier for those trying to infiltrate.
Artificial intelligence for cyber crimes accelerates the development of new malware
Next to these concrete episodes, some scenarios that we could define as “experimental” also emerge. ESET researchers identified a ransomware prototype called Promptlockwhich uses the AI to constantly generate new harmful code. The AI -based malware is able to select the most sensitive files, copy and encrypt them (i.e. make them illegible without a key). Although it has not yet been widespread on a large scale, it shows how the AI can greatly accelerate the creation of new malware. Not by chance Anton Cherepanovone of the researchers who helped to find Promptlock, explained:
With the help of artificial intelligence, launching sophisticated attacks has become considerably easier. A well -configured model is now sufficient to create a complex and self -adapting malware. If adequately implemented, these threats could seriously complicate the detection and make the work of the defenders of cyber security significantly more demanding.
