67% of Banned Accounts Used AI for Cyberattacks, Says Anthropic

Anthropic, a leading AI company, announced that 67% of accounts it banned for policy violations between March 2025 and March 2026 utilized AI in preparing cyberattacks. Out of 832 analyzed accounts, 560 leveraged AI tools for tasks like writing malware or exploiting vulnerabilities, according to the company’s findings released on June 4, 2026.

The data highlights a troubling trend: AI is enabling a broader range of actors to execute sophisticated attacks. For instance, Anthropic noted that 6.5% of the banned accounts used AI not just for preparation but for deeper stages of the attack lifecycle, such as “lateral movement”—a phase where attackers expand their access after breaching a system. Traditionally, such advanced techniques required significant technical expertise, but AI has lowered the barrier for less-skilled cybercriminals.

The prevalence of AI-driven cyberattacks coincides with a surge in crypto-related hacks. In April 2026, stolen crypto value hit $629.7 million, the highest monthly total since February 2025, with analysts partly attributing this spike to AI’s growing role in identifying and exploiting vulnerabilities. Manuel Aráoz, founder of OpenZeppelin, recently declared that “all of DeFi is unsafe” due to AI’s capacity to uncover flaws in smart contracts.

Anthropic’s report also pointed to a rise in the threat level posed by AI-enabled attackers. In the first half of its study, 33% of banned accounts were classified as medium or high risk. By the second half, this figure had jumped to 56%. The company cited an example of a Chinese state-sponsored group using an autonomous AI model to conduct an attack in November 2025. The AI carried out complex tasks, including exploiting vulnerabilities and stealing credentials, with minimal human input.

These findings arrive at a pivotal moment for Anthropic. On May 28, the company closed a $65 billion funding round, valuing it at $965 billion—the highest private valuation for an AI firm to date. This funding positions Anthropic for a potential IPO as it scales its operations and continues to develop advanced AI models. Its latest product, Mythos, is set to launch soon, promising powerful cybersecurity capabilities capable of identifying thousands of software vulnerabilities. However, these capabilities also raise questions about the potential misuse of such technology.

The report underscores the dual-edged nature of AI development. While advancements like Anthropic’s Claude models are celebrated for their ability to audit code and enhance enterprise security, they also highlight how malicious actors can weaponize the same tools. Google researchers recently uncovered what they believe to be the first instance of AI being used to craft a zero-day exploit that bypassed two-factor authentication—a sobering reminder of the risks associated with increasingly capable AI systems.

Anthropic’s findings serve as a call to action for regulators, enterprises, and developers to address the growing role of AI in cyber threats. With AI continuing to evolve and its deployment expanding, the stakes for cybersecurity have never been higher.

Image source: Shutterstock