Anthropic Mythos AI model reportedly poses unprecedented cybersecurity risks
Anthropic warned in a leaked blog post that its upcoming Mythos AI model can exploit software vulnerabilities at unprecedented speeds, faster than human hackers. The company and other AI developers are allowing select organizations to test the models and are warning officials, while experts caution that AI agents could enable large-scale cyberattacks against governments, corporations, and critical infrastructure worldwide.
What happened
Anthropic published a draft blog post warning that its upcoming Mythos AI model can exploit software vulnerabilities at an unprecedented pace. The post, which leaked, stated: 'Although Mythos is currently far ahead of any other AI model in cyber capabilities, it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.' Anthropic said the leak resulted from human error in its content management system.
OpenAI similarly warned in December that its upcoming models posed a 'high' cybersecurity risk. Anthropic is allowing select organizations to test Mythos ahead of release to improve their defences against AI-driven exploits, and is privately warning government officials about potential large-scale cyberattacks enabled by the model.
Reported impact
Cybersecurity experts warn that AI-powered cyberattacks are escalating because AI agents can find security weaknesses faster than human hackers. A single AI agent could scan for vulnerabilities and potentially exploit them faster and more persistently than hundreds of human hackers. Cato Networks founder Shlomo Kramer stated: 'The agentic attackers are coming. This is a watershed event in the history of cybersecurity.'
Index analysis
The threat reportedly affects governments, corporations, and critical infrastructure worldwide. Examples cited include a Russian-speaking hacker reportedly using multiple AI tools including Anthropic's Claude and China's DeepSeek to hack over 600 devices in 55 countries in January, and Claude reportedly being used in attacks on Mexican government agencies in February where sensitive voter and tax data were leaked.
Governance controls
Anthropic and other AI development centres are conducting tests on their models with select organisations to prepare for large-scale AI-based cyber threats. Experts highlight how AI-based cybersecurity tools will be essential in combating the development of autonomous AI attacks, with AI enabling faster monitoring and automated patching.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.