Safety of using AI remains a question mark

India's CERT-In has issued a critical advisory, CIAD-2025-0013, addressing the rising perils of generative AI. This warns of susceptibilities within AI models, demanding robust mitigation strategies. 

As AI permeates sectors like healthcare and finance, it also breeds intricate cyber threats, exploiting weaknesses in training data and model design.

The advisory delineates attack vectors, including data contamination, where manipulated training data yields flawed AI outputs. Adversarial attacks subtly alter inputs, deceiving AI systems. Model inversion and theft allow extraction of training data and replication of proprietary AI, threatening intellectual property.

Prompt injection embeds malicious commands, manipulating AI responses. Hallucination exploitation spreads false information, while backdoor attacks implant hidden triggers for later breaches. These threats necessitate stringent security protocols.

CERT-In provides best practices for secure AI utilization. Users must verify AI tool authenticity, avoiding unproven applications. Personal data should not be shared with AI services, particularly cloud-based ones. Access rights should be meticulously managed to prevent data leaks.

Reliance on AI for absolute accuracy is discouraged, as biased training data can produce unreliable outputs. AI usage should be confined to its intended scope, not replacing human judgment in critical areas. Robust cybersecurity practices, like multi-factor authentication, are essential.

Anonymous accounts and anonymized data protect personal information. Vigilance against plagiarism and AI-generated scams is vital. Deepfakes and fraudulent content demand skepticism.

As AI evolves, so do its associated risks. Awareness and proactive security measures are paramount. Organizations and individuals must conduct risk assessments and stay informed. Responsible AI adoption is crucial for preserving trust and security in an AI-driven landscape.