Advanced AI models high on priority: IT secy, Pg17

Key Highlights:

• The Indian government is prioritizing access to Mythos, an advanced AI model developed by Anthropic, for detecting software vulnerabilities.
• IT Secretary S Krishnan stated that negotiations are underway with the U.S. and relevant companies to secure access to Mythos and similar frontier AI models.
• In the interim, the Indian Computer Emergency Response Team (CERT-In) has established a "sandbox" using alternative AI models with 60-70% of Mythos's capabilities to identify and fix vulnerabilities.
• The government emphasizes the need for on-premises deployment for highly sensitive code, rather than relying on overseas cloud services.

Detailed Insights:

• Mythos, formally known as Claude Mythos Preview, is an AI model designed for autonomous security research, capable of finding and exploiting zero-day vulnerabilities.
• This advanced AI can reason over source code, form hypotheses about exploitable conditions, and generate working proof-of-concept exploits.
• Anthropic launched Project Glasswing to deploy Mythos defensively, partnering with various technology companies to find and patch security flaws in critical software.
• The Indian government's pursuit of such models aligns with its broader National Strategy for Artificial Intelligence, developed by NITI Aayog, which aims to position India as a global AI leader.
• This strategy focuses on leveraging AI for economic growth, improving social outcomes, and addressing national challenges across sectors like healthcare, agriculture, and education.
• The use of alternative AI models in a sandbox environment by CERT-In allows for continuous testing and development of secure workflows while formal access to Mythos is being negotiated.

Key Concepts Involved:

• Artificial Intelligence (AI): The simulation of human intelligence processes by machines, especially computer systems.
• Frontier AI Models: The most advanced and capable AI systems currently available, often characterized by their large scale and broad applicability.
• Zero-day Vulnerability: A software flaw unknown to the vendor, for which no patch exists, making it highly exploitable by attackers.
• Sandbox Environment: An isolated testing environment that enables users to run programs or open files without affecting the application, system, or platform on which they run.