CERT-In proposes AI security audit guidelines with 4 risk levels
CERT-In has proposed new AI security-audit guidelines for AI systems to keep things safer and more transparent.
The guidelines sort AI into four risk levels, low to critical, and now even big names like OpenAI could face closer audits.
The main idea? Make sure AI tools don't get more access than they need.
CERT-In proposes kill switches, continuous monitoring
High-risk agentic AI systems would need a "kill switch" so it can be shut down fast in emergencies.
Critical-risk AI systems would face continuous monitoring and independent security audits conducted by organizations empanelled by CERT-In.
Audits aren't just about hacking: they'll also check for stuff like fake info (AI hallucinations) and bias.
Plus, any content made by AI is expected to use labeling or watermarking to identify synthetic content, depending on the nature and context of its use, and companies must report security issues within 6 hours.
Basically, India wants every part of the AI process covered from start to finish.