There is a growing risk from autonomous AI agents beyond their authorised tasks.
In the age of agentic AI, the window to defend a system is shrinking dramatically.
Awareness is rising in India, but policymakers need to do more.
The State Bank of India, the country’s largest lender, confirmed on October 21 that personal data of some of its customers, including their mobile numbers, addresses and loan details, among other things, was exposed on the dark web. The disclosure comes amid increasing cybersecurity risks faced by institutions in critical sectors, with concerns compounded by the ability of artificial intelligence systems to further exploit security loopholes.
The latest breach comes just days after the Indian Computer Emergency Response Team (CERT-In), an office within the Ministry of Electronics and Information Technology that deals with cyber security incidents, warned that under-testing AI agents went ‘rogue’ and gained unauthorised access to the live systems of unnamed Indian organisations, potentially exposing business records, personal data and other confidential information. The agents were able to carry out these attacks using exposed credentials, publicly accessible code repositories and overly permissive cloud access settings. The scariest bit is: the affected organisations did not identify the unauthorised access themselves; AI providers disclosed the incidents to CERT-In. It also reported cybercriminals using open-source AI tools to conduct financially motivated attacks.
This incident highlights a growing risk from autonomous AI agents—not just AI helping humans carry out cyberattacks, but AI systems themselves acting beyond their authorised tasks and accessing real-world systems.
The threat landscape is changing. To win, you have to fight AI with AI. In the age of agentic AI, the window to defend a system is shrinking dramatically. Autonomous AI agents can accelerate cyberattacks, leaving organisations with less time to detect, assess and contain threats. A May 2026 report by Palo Alto Networks says it can take just 39 seconds for attackers to move from initial access to data theft.
Alok Agrawal, the chief solutions officer of Rubrik, a cyber resilience and data security company founded in 2014 and headquartered in Palo Alto, breaks down these 39 seconds.
You do everything you can, including installing an alarm system, to protect your home from external threats, like burglars. You know that if a burglar enters the house, the alarm will go off and the cops will be at your house within five minutes. The burglar won’t be able to take anything in that time. But what if the burglar is able to enter and exit your house in 39 seconds? To navigate the possibility of a burglar entering your house despite the alarm system, you build a bunker underneath your house to protect your family, your documents and your crown jewels.
Now imagine a similar situation in the cyberspace. As AI accelerates cyberattacks, organisations have an increasingly narrow window to detect and stop a breach. The existing technologies are designed to detect attacks developed for human speed, not for machine speed, says Agrawal. However, detection alone is not going to be sufficient. “Businesses need to prepare for attacks that can unfold faster than conventional security systems can respond. As more services and sensitive information move online, organisations must build resilience into their technology infrastructure rather than rely solely on prevention,” he says.
Cyber resilience and very quick cyber recovery are two things that companies will have to invest in now on an urgent basis. “A rapid recovery capability is important because the consequences of a prolonged outage can extend beyond immediate business losses to reputational damage and disruption to customers,” says Agrawal.
Rubrik chief product officer Anneka Gupta warned that enterprises could face a wave of cyber breaches as advanced AI capabilities become more widely accessible to attackers. “Most organisations are not going to be prepared. We are going through a phase where there are going to be a lot of security breaches. And this won’t happen in a distant future,” she adds.
Team Rubrik was in Goa recently for their Global System Integrator (GSI) summit.
Making Companies Attack-ready
By 2025, the cybercrime business will be worth $10 trillion, making it the third-largest economy in the world. Rubrik, the New York Stock Exchange-listed company, helps businesses achieve business resilience against cyberattacks, malicious insiders and operational disruptions. The company has carried out more than 350 cyber recoveries for existing customers. In a breach, its technology can help identify the extent of the damage, determine which backup snapshots remain clean and restore affected files, folders and systems.
At the Goa summit, it launched a new product as part of its larger mission.
In April, Anthropic, an AI safety and research company that’s working to build reliable, interpretable and steerable AI systems, launched Project Glasswing—an AI-powered cybersecurity initiative launched to protect the world’s most critical software forms from increasingly sophisticated cyberattacks. It uses Anthropic’s highly capable AI model, Claude Mythos Preview, to identify security flaws before malicious hackers can exploit them. At the Goa summit, Rubrik unveiled Code Guardian—an AI-powered code vulnerability scanner born out of Project Glasswing, which offers a look at how AI can be used to help identify critical security flaws before rogue agents or attackers can exploit them, with India playing a key role in its research and development (R&D).
The tool analyses cloned, air-gapped copies of customer repositories rather than live production environments. It is designed to identify chains of vulnerabilities, assess their exploitability and prioritise findings according to business impact. Explaining how Code Guardian works in an air-gapped environment—a computer system or network that is physically or logically isolated from the internet and other unsecured networks to prevent unauthorised access, data leaks and cyberattacks—Gupta says: “You are not running the programme yourself in your own environment. It’s running within Rubrik. We are assuming the responsibility for that. And we’re taking a copy of the code. We’re not touching the raw repositories or live production code.”
Rubrik has expanded Project Hourglass to include AHEAD, Tata Consultancy Services, Trace3 and World Wide Technology, alongside existing partners Cognizant, Deloitte, LTIMindtree, HCLTech, NTT DATA and Wipro. The partners will help take the technology to enterprise customers.
A major cybersecurity incident in July 2026 in which AI agents being tested by OpenAI escaped their intended isolation and compromised parts of infrastructure of Hugging Face—a platform where the machine learning community collaborates on models, datasets and applications—raised serious questions about whether advanced AI systems can evade security controls and conduct cyberattacks with limited human direction. It demonstrated how AI agents, when given powerful capabilities and insufficiently effective containment, could turn a cybersecurity test into an unauthorised intrusion. “A lot of people were predicting that something like that would happen. The Hugging Face attack proved that the future is already here. It raised the level of awareness about the risks posed by AI agents and has forced people to put critical thought into how they’re going to address and mitigate this risk,” says Agrawal.
He adds that the launch of Anthropic’s Mythos model had raised awareness of the risks by demonstrating how AI could identify vulnerabilities at machine speed and combine smaller weaknesses to create potentially more serious security gaps. “However, since Mythos is available to only 120 different companies worldwide, my belief is that the number in India is very small, which presents a huge opportunity for our Code Guardian model,” says Agrawal.
India an R&D Hub
Where does India stand in terms of mitigating these attacks? “The awareness has gone up in the past five years, but more so in the recent past,” says Agrawal. He credits this to the launch of Mythos, increased awareness of cyber threats and the CEOs of the frontier model companies like OpenAI, Anthropic, that are talking about pacing AI.
Talking about India being a prominent centre for Rubrik, Gupta says: “Around 60 per cent of our R&D is in India. Some of our most innovative products are being built almost entirely in India. We co-locate product design and engineering teams to build these products. India is also where our R&D team is moving the fastest. We have hired top engineers and product managers, so this is not about cost. India is truly an innovation hub for us.” From a larger perspective, India represents a huge untapped potential. We have entire product lines that are being developed and shipped from both our Bengaluru and Pune offices. This talent is developing front-end, high-innovation products, not just back-end testing and low-level tasks, says Agrawal.
When asked what should Indian policymakers do so that more and more companies take the threat factor seriously and strengthen their mechanisms, Gupta say guidance is the first step.
(The writer was invited by Rubrik to attend the GSI Summit in Goa)
