Cybersecurity experts have begun warning that the threat of AI-powered attacks is growing after Anthropic developed the Mythos model that can find security weaknesses faster than human hackers.
This threat is impacting governments, businesses, and critical infrastructure around the world, raising concerns about how to defend against AI exploits in real time.
According to a leaked blog post, AI developer Anthropic has warned that its next Mythos model will perform automatic software vulnerability scanning and exploitation at an unusually fast rate. OpenAI issued a similar warning in December about future models that pose high cybersecurity risks.
Shlomo Kramer, founder and CEO of Cato Networks, commented on this: “Agent attackers are coming. This is a momentous event in the history of cybersecurity.” Experts say one AI agent can beat hundreds of human hackers by continuously scanning, testing, and exploiting vulnerabilities.
Russian-speaking hackers used multiple AI tools, including Anthropic’s Claude and China’s DeepSeek, to hack more than 600 devices in 55 countries in January. Amazon Web Services’ security research team says this was done using AI tools. According to Eyal Sela, director of security for threat intelligence at Gambit, AI has made it easier for attackers to perform complex hacks, giving them “superpowers” regardless of skill.
Claude was also used in an attack on Mexican government agencies in February, reportedly leaking sensitive voter and tax data.
AI can also be used to improve cybersecurity. Joe Lin, co-founder and CEO of Twenty, commented, “An attacker only needs one way to get in, but a defender needs to cover all sides. AI allows for faster monitoring and automated patching, but humans must control the outcome.”
Experts emphasize how AI-based cybersecurity tools are essential to countering the development of autonomous AI attacks.
Anthropic and other AI development centers test selected organizations and models in preparation for large-scale AI-based cyber threats.
