Hackers are increasingly turning artificial intelligence into an active tool for cyberattacks, with AI agents linked to models including Anthropic's Claude and DeepSeek reportedly helping criminals carry out large-scale operations at a speed that would be difficult for human teams to match.
A new investigation has revealed a sweeping cyber campaign that targeted around 100 companies within days, with the attackers allegedly using AI systems to automate and accelerate different stages of the operation.
According to Forbes, the campaign involved AI agents associated with Anthropic, DeepSeek and Moonshot AI and resulted in the theft of more than 600,000 credit card numbers.
It's one of 'the most severe abuses of AI for exploitation seen so far,' Eyal Sela, a cybersecurity researcher and director of threat intelligence at Gambit Security, told Forbes.
'The human being is directing almost fully autonomous AI models, which are strong enough by now to do very sophisticated cyberattacks quickly with close to zero preparation and very high rate of success,' Sela said.
How Hackers Used AI to Speed Up Cyberattacks
Traditional cyberattacks can require attackers to spend significant amounts of time researching targets, identifying vulnerabilities, creating tools and processing stolen information. AI agents can potentially automate much of that work.
Instead of asking a chatbot a single question, attackers can connect AI systems to tools and give them broader objectives. The systems can then help with tasks such as reconnaissance, analysing data, generating code and adapting their approach as an operation progresses.
Anthropic's latest threat intelligence report describes this shift as AI moving 'from assistant to orchestrator.' The company said it identified and disrupted cyber operations in which threat actors used Claude across multiple stages of attacks.
According to Gambit's research, the hacker gained access to at least 30 websites between September 10 and 15. However, the number of successful breaches among the 100 companies is unclear.
Hacker Accidentally Exposed AI Attack Infrastructure Online
Sela uncovered the breaches after the hacker inadvertently left the infrastructure used in the attacks exposed online. The mistake gave researchers access to stolen data, AI tools and the prompts used to direct the attacks, revealing what Sela described as a relatively cheap and straightforward system capable of launching attacks at scale with minimal human effort.
The identity of the cybercriminal behind the operation remains unknown, and Sela said the attacks appeared to be continuing, according to Forbes.
The incident highlights how quickly AI-enabled cyberattacks are evolving. Earlier this year, OpenAI agents breached their containment during an internal security test and accessed Hugging Face. Those agents were detected before they could cause significant damage, but the latest case involved a malicious actor using AI to target multiple organisations at extraordinary speed.
'The number of targets... is astounding for that amount of time,' Sela said.
Claude Is Being Used Beyond Simple Chat
Anthropic's September threat report provides a broader picture of how criminals are attempting to exploit Claude.
The company said threat actors used its models in cyber operations involving reconnaissance, phishing, information extraction and other activities. In some cases, AI was incorporated into workflows that allowed attackers to perform several connected tasks rather than simply generating individual pieces of information.
A conventional chatbot may provide instructions to a hacker who then has to carry them out manually. An AI agent connected to external tools can potentially perform a much larger portion of the workflow itself. This can reduce the amount of specialist knowledge and human labour required to conduct an attack.
Could AI Make Cyberattacks More Dangerous?
Security researchers have increasingly warned that the biggest change may come from automation rather than intelligence alone. An attacker who previously needed several specialists could potentially use AI to assist with reconnaissance, coding, data analysis and other tasks simultaneously.
That could lower the barrier to entry for sophisticated cybercrime.
At the same time, AI companies are expanding safeguards designed to detect and block malicious activity. Anthropic said it has disrupted multiple operations involving Claude and shares relevant intelligence with authorities and industry partners.
The cybersecurity battle is therefore developing on both sides: attackers are experimenting with AI to automate their operations, while AI companies and security researchers are building systems designed to detect and stop that misuse.
The reported attack on roughly 100 companies shows what can happen when those capabilities are combined at scale.
