In an expanding digital universe, AI has assumed great significance. For countries like India, the impact can be huge. An EY report predicts that generative AI has the potential to add a massive $1.2-1.5 trillion to India’s GDP over the next seven years. However, every technology can be used for both good and bad intentions. As organisations embrace the positive impact of AI-driven innovations, they must also confront a silent threat called AI poisoning.
The modus operandi: AI poisoning involves attackers corrupting training data, causing AI to make biased or wrong assumptions. “The consequences can be catastrophic,” says Rohan Vaidya, area vice-president, India & SAARC, CyberArk. “Picture a self-driving car that has been trained on manipulated road signs embedded within its database leading it to cause accidents that may even lead to loss of lives. Similarly, malicious actors could embed fraudulent transactions into historical financial data would result in loans being given to undeserving applicants and cause financial losses. There are also possibilities of chatbots being trained on poisoned data filled with inflammatory language which could then become a tool for spreading misinformation and discord,” he cautions.
A growing concern: AI poisoning poses significant risks to Indian enterprises, potentially causing financial losses, reputational damage, and security breaches. Malicious actors can exploit this vulnerability to launch targeted attacks, warns Vaidya. “Poor choices made by AI systems might increase financial risks by resulting in unsuccessful transactions, expensive errors made in a variety of processes. Furthermore, organisations’ reputations might be damaged and their products and services would lose credibility if customers believe that AI systems are manipulable,” he adds. Additionally, AI poisoning could lead to ethical and legal issues, potentially violating privacy and data protection laws like the Personal Data Protection Bill.
The remedy: One of the foundational pillars in defending against data poisoning is sourcing the right type of data. Organisations must scrutinise data sources rigorously, prioritising authenticity, reliability, and relevance.
Building adaptive AI models is another key strategy in combating data poisoning. These models are designed to dynamically adapt to changing data distributions, ensuring continued relevance and accuracy over time. Continuous monitoring and regular audits are also essential components of a proactive defense against data poisoning, Vaidya summarises.