The threat actor, which Anthropic said was likely to be a Chinese state-sponsored group, “succeeded in a small number of cases,” according to a blog post [File]
| Photo Credit: REUTERS
AI company Anthropic disclosed the details of a “highly sophisticated AI-led espionage campaign,” or AI cyberattack where the threat actor manipulated the company’s Claude Code tool to attempt infiltration of around thirty global targets.
The threat actor, which Anthropic said was likely to be a Chinese state-sponsored group, “succeeded in a small number of cases,” according to a blog post published on Thursday (November 13, 2025).
The company detected suspicious activity in mid-September this year, as the attackers used AI’s “agentic” capabilities. Over the following ten days, Anthropic said it investigated the operation, banned the accounts it identified, notified the affected entities, and worked with the authorities.
Anthropic disclosed that the attack targeted large tech companies, financial institutions, chemical manufacturing companies, and government agencies.
The AI company backed by Google and Amazon noted that the attack was significant as the threat actor used AI to perform 80-90% of the campaign and achieved an attack speed of thousands of requests per second, which would be unthinkable for human cyber criminals.
The AI-powered attack was not perfect due to hallucinations and other inaccuracies. However, the security breach showed how the more autonomous nature of AI agents makes it possible for hackers to use them to execute advanced cyber attacks and not just source advice.
“We believe this is the first documented case of a large-scale AI cyberattack executed without substantial human intervention. It has significant implications for cybersecurity in the age of AI agents,” posted the company on X on November 13.
“Vibe hacking,” in which attackers use AI to generate malicious code, is a rising risk, but Anthropic stressed that the latest attack was a step above even that. The threat actors were able to “jailbreak” Claude and tell it that it was an employee at a real cybersecurity firm.
“When sophisticated cyberattacks inevitably occur, our goal is for Claude—into which we’ve built strong safeguards—to assist cybersecurity professionals to detect, disrupt, and prepare for future versions of the attack. Indeed, our Threat Intelligence team used Claude extensively in analyzing the enormous amounts of data generated during this very investigation,” said Anthropic, suggesting that the same abilities that allow Claude to be used for cyberattacks could also enable it to help in cyber defence.
Published – November 14, 2025 01:33 pm IST
