Hyderabad: The city’s police commissioner, VC Sajjanar, has issued a warning about potential information leaks from AI chatbots due to a tactic called “prompt injection.”
He explained that cybercriminals can feed harmful instructions into AI systems, compromising sensitive corporate and customer data. Businesses are increasingly using AI for its efficiency in handling customer inquiries, speeding up tasks, and cutting costs. However, the risk of “prompt injection” looms large.
In AI, instructions given to execute tasks are termed “prompts.” Criminals exploit these by inserting harmful prompts, which deceive the AI into leaking confidential information.
Sajjanar emphasized that a “prompt injection attack” can trick AI into exposing internal files, customer databases, and private system details that should remain secure.
This method poses a significant threat to data security. Many organizations integrate AI with their key systems, such as CRM data and financial records, making them vulnerable. An ill-intentioned prompt could inadvertently reveal this information to hackers.
To combat these risks, Sajjanar recommends implementing “prompt guardrails” and adopting a multi-layered security strategy. He urges companies to train their teams on AI safety and set stringent controls at both model and system levels.
He further advises conducting regular security audits to manage data access effectively. Sajjanar warns that without these precautions, organizations risk operational disruptions and loss of valuable data to cybercriminals.
