Rating the threat from Anthropic’s Frontier AI model ‘Claude Mythos’ as “high severity” , the CERT-In (Indian Computer Emergency Response Team) of India’s IT Ministry called for heightened vigilance, personnel training and incident response plan for organizations and MSMEs.
In a detailed advisory issued to organizations and MSMEs on Sunday night, the CERT-In said that Claude Mythos AI model has reported capabilities to “autonomously discover security vulnerabilities in widely used software, analyze source code, plan and chain together multi stage attacks, and carry out simulations of compromising enterprise networks from end to end,” .
CERT-In, the national incident response center for major computer related incidents in India, also observed that previously such attacks at the speed and scale capability of Claude Mythos required teams of skilled human experts.
“Potential impacts include unauthorized access, service disruption, data exfiltration, identity compromise, financial fraud, impersonation, persistent compromise of operational environments, and cascading compromise of interconnected systems and services,” read the advisory.
The advisory by CERT-In comes days after India’s Finance Minister Nirmala Sitharaman cautioned banks on the cybersecurity risks posed by frontier AI models like Claude Mythos and called for heightened vigilance and need for institutionalized defense mechanisms against automated bad actors.
India now joins the U.S., U.K. and Singapore as countries that have issued advisories and cautioned banks and financial institutions against risks posed by AI model Claude Mythos, that was unveiled late March this year by Anthropic.
The US based AI startup, Anthropic, has announced that it will not release Claude Mythos for the general public owing to the cybersecurity risks and would rather allow MNCs, tech giants and countries to assess their cyber strengths using a preview model of their Frontier AI.
Post the unveiling of Claude Mythos Preview model, U.K. based AI Security Institute (AISI) ran a test and published its results stating that the AI model was found to be the “most capable”. India’s CERT-In has referred to the same report while issuing the advisory against Mythos.
CERT-In advisory has called for Zero Trust Network Architecture (ZTNA) for organizations that includes applying “Zero Trust principles by treating every access request as untrusted by default” and “ granting users and systems only the minimum access they need to do their jobs”, and “assuming that a breach may already have occurred.”
CERT-In has also advised for multi-factor authentication (MFA), advanced micro-segmentation, urgently treating critical patches while DDoS protection, continuous security monitoring, and rapid incident detection must become standard practice to prevent automated reconnaissance, credential theft, and AI-driven exploit campaigns.
Also Read: Claude Mythos: Imminent Threat or Marketing Hype by Anthropic?
