Agentic AI is a tool used by both offenders and defenders.
The cyber threat landscape is rapidly transforming into a virtual battleground that reads like a page out of a sci-fi novel. With GenAI-powered Agents sprouting, the clash between the defenders and hackers have moved on to a new orbit. From tech giants like Google to cybersecurity firms are deploying intelligent Agents to keep tabs on the ever-evolving threat actors.
“Our AI agent Big Sleep helped us detect and foil an imminent exploit. We believe this is a first for an AI agent – definitely not the last – giving cybersecurity defenders new tools to stop threats before they are widespread,” Sundar Pichai, Chief Executive Officer of Alphabet, the parent organisation of Google, has said in a post on X.
The tech major, which is among the frontrunners in AI research, launched ‘Big Sleep’, an Agentic AI solution that helps defenders in tackling advanced cyber threats.
Most notably, it detected an SQLite vulnerability (CVE-2025-6965) known only to threat actors, allowing Google to predict its imminent exploitation and cut it off beforehand.
Google believes this marks the first time an AI agent has been used to directly foil exploitation efforts in the wild.
Ankit Sharma, Senior Director and Head – Solutions Engineering at Cyble, explains that this software, capable of planning, reasoning and acting autonomously, is now on the frontlines. Agentic AI is a tool used by both offenders and defenders. While attackers wield it to probe networks, customise phishing at scale and exfiltrate data, defenders use it to hunt threats and prevent these very attacks.
Pankit Desai, CEO & Co-founder of Sequretek, confirms this capability. “Our AI agents understand the inbound communications to the network and if the impact to the business is low but the impact to security is high, then the agent automatically blocks the communication,” he says.
Agentic AI can theoretically spot emerging vulnerabilities and even pre-emptively block or patch an unknown exploit kit under testing.
However, the cybersecurity companies, for now, are focussing on spotting and blocking incoming threats, including the AI-powered ones, and not proactively chasing them down.
“Our platforms like Percept CTEM and Percept Compliance Manager offer proactive defence, but always with explainability and control at the core. As attackers deploy their own AI, the future of security will rest on defensive AI coalitions, rapid adaptability and responsible automation,” Desai said.
Sharma felt that the most effective approach for using agentic AI defensively blends autonomous capabilities with transparent human oversight.
“Agentic AI can handle the scale and speed of modern attacks, but it needs clear, implementable and constantly updatable guardrails. Spotting and thwarting attacks before they mature isn’t just a technological problem, as we have learnt over the past decades of cybersecurity evolution — it’s an organisational and ethical one,” he said.
More Like This
Published on July 19, 2025