EY India’s Raghavendra BV urged businesses to adopt real-time monitoring, threat modelling and tailored kill switches as AI makes cyberattacks faster.
Key Highlights:
- AI can compress cyberattack planning and execution from months to minutes, an EY expert said.
- Raghavendra BV estimated that attackers’ speed has increased almost 100 times.
- Businesses should involve cybersecurity experts while developing AI agents.
- Runtime protection can help organisations monitor agents’ actions continuously.
- Kill switches must be tailored to an agent’s tasks and normal behaviour.
- EY India unveiled AI solutions for finance, cybersecurity and supply chains.
New Delhi, November 19: Cyberattacks that once required months of planning can now be launched within minutes using artificial intelligence, making it essential for businesses to build security safeguards into AI systems from the outset, an EY India cybersecurity expert said on Thursday.
Speaking at the EY Technology Dialogue, Raghavendra BV, Partner and Deputy Leader of Cybersecurity Consulting at EY India, said AI had sharply accelerated attackers’ ability to target organisations.
He said attacks, including state-sponsored operations aimed at critical infrastructure such as power grids and railways, previously took months to conceptualise, plan and execute. With AI, some could now be carried out in minutes, he said, estimating that attackers’ speed had increased almost 100 times.
Security must begin during development
Raghavendra warned that traditional detection methods could prove too slow against rapidly moving threats. Organisations need AI-enabled systems capable of identifying attacks and responding within milliseconds, he said.
He identified the deployment of AI without adequate guardrails as a major risk and urged companies to involve cybersecurity experts during agent development.
Adding safeguards only after something goes wrong is less effective than embedding security at the beginning of an organisation’s AI journey, he said.
He advocated AI threat modelling: assessing what a system can access and do, how it might be attacked or misused, and which safeguards it needs before deployment. Such assessments are relevant across industries, rather than only regulated sectors such as banking, he added.
Continuous monitoring of AI agents
Raghavendra said gateways and language model firewalls provide safeguards at entry points, but businesses also need runtime protection to monitor agents while they operate.
He described this as software running alongside an agent that reports its activities to a central control layer. Organisations can then analyse behaviour, detect unusual actions and activate a kill switch when necessary.
However, he cautioned that kill switches cannot rely on a universal, ready-made rule set. Companies must understand each agent’s tasks and establish its normal behaviour before identifying meaningful anomalies.
For example, an unexpected attempt to access a system outside an agent’s usual pattern could warrant intervention.
EY India also unveiled AI solutions for finance functions, agentic cybersecurity and supply chain forecasting, planning and operational efficiency. The company said its cybersecurity capabilities are supported by more than 5,000 cybersecurity and digital risk professionals in India.









