Government has some important tips for using ChatGPT, Gemini, Grok and other AI tools safely

AI

Indian Computer Emergency Response Team (CERT-In) has issued an advisory outlining best practices for using generative AI tools like ChatGPT, Gemini, and Grok safely. The advisory highlights the risks associated with AI applications, including data poisoning, adversarial attacks, and model stealing, while also providing guidelines to ensure responsible use of AI tools.

CERT-In warns that generative AI models are susceptible to various security threats. These include:

Data poisoning: Malicious actors can manipulate training data to mislead AI models, causing them to generate biased or incorrect outputs.

Adversarial attacks: Attackers can subtly modify inputs to trick AI into producing false responses.

Model inversion: Hackers may extract sensitive details from AI training data.

Model stealing: Threat actors can replicate AI models by continuously querying them.

Prompt injection: Malicious inputs can bypass content filters and exploit AI responses.

Hallucination exploitation: Attackers can misuse AI-generated misinformation to spread false narratives or scams.

Backdoor attacks: Hidden triggers within AI models can lead to unexpected, potentially harmful behaviors.