US artificial intelligence company Anthropic, the developer of the chatbot Claude, has revealed its technology has been misused by hackers to carry out cyber attacks, large-scale fraud, and even international sanctions breaches.
The company disclosed that criminals exploited its tools “to commit large-scale theft and extortion of personal data”, while North Korean operatives used Claude to help them fraudulently secure remote jobs at major US companies.
Anthropic said it has been able to disrupt these activities, report them to the authorities, and strengthen its systems to improve detection. But the cases highlight how quickly malicious actors are adapting powerful AI tools to their advantage.
AI and “vibe hacking”
One of the most striking incidents involved so-called vibe hacking – where attackers used AI to generate malicious code targeting at least 17 organisations, including government bodies.
According to Anthropic, this was not limited to producing software. Hackers relied on Claude to help direct the entire campaign, from choosing what data to steal to drafting extortion messages tailored to victims. The chatbot even suggested ransom amounts.
The company described this as AI being used “to what we believe is an unprecedented degree”. The incident has raised concerns that as AI grows more capable, it may enable attacks that are broader, faster and more psychologically manipulative than ever before.
The rise of agentic AI
Much of the anxiety stems from the development of Agentic AI – models designed to act more autonomously, taking decisions without step-by-step human instruction.
Advocates argue this autonomy could power everything from advanced digital assistants to business process automation. But in cyber crime, the same capabilities could allow attackers to scale operations and reduce the time it takes to exploit vulnerabilities.
“The time required to exploit cybersecurity vulnerabilities is shrinking rapidly,” said Alina Timofeeva, an adviser on cyber-crime and AI. “Detection and mitigation must shift towards being proactive and preventative, not reactive after harm is done.”
North Korean operatives and remote jobs
Anthropic also revealed its AI was used by North Korean operatives seeking to infiltrate US firms.
Remote working has long been seen as a potential weak spot in cyber security. North Korean groups have previously posed as overseas developers to gain access to company systems. But Anthropic says the use of AI to generate convincing CVs, cover letters, and communications marks “a fundamentally new phase” in these schemes.
Once hired, the workers relied on AI to translate messages and assist with coding tasks, according to the company.
Anthropic’s threat intelligence report said the use of AI created “a new paradigm” that allows less skilled operatives to overcome cultural and technical barriers, making it easier for them to obtain work with international employers. This not only risks exposing company systems to infiltration, but may also result in firms inadvertently breaching international sanctions by paying North Korean nationals.
A new phase, not a new crimewave
Despite the alarming revelations, experts caution against assuming AI has triggered a wholly new surge in cyber crime.
Many ransomware attacks still rely on tried-and-tested techniques such as phishing emails and exploiting unpatched vulnerabilities. However, researchers warn that AI is lowering the barriers, giving attackers the ability to operate at speed and scale with less technical expertise.
As Jacob Klein, Head of Anthropic’s Threat Intelligence Team, explained:
“If you’re a sophisticated actor… now, a single individual can conduct [attacks], with the assistance of agentic systems.” (The Verge)
Balancing innovation and risk
Anthropic’s disclosures underline a growing challenge for policymakers, regulators and businesses. On one hand, AI promises efficiency, innovation and productivity. On the other, the same tools can be turned against individuals, companies and even governments.
As AI becomes more autonomous, the line between human-led and machine-led attacks may blur further. Security specialists say the priority now is to move beyond reacting to breaches, and towards anticipatory defence strategies that account for how criminals are already experimenting with these tools.
For Anthropic, the incidents are a stark reminder that even companies at the forefront of AI development cannot fully predict how their technology will be used once released into the wild. For organisations, the lesson is clear: cyber security in the age of AI will demand not just stronger technical safeguards, but a shift in mindset towards proactive, AI-aware defence.
