As AI agents become increasingly capable of finding and exploiting software flaws, whether through unintended behaviour or on the behalf of malicious actors, Indian enterprises need to ensure that their codebases are secure before deployment because the stakes are much higher now.The Indian Computer Emergency Response Team (CERT-In) on Thursday, October 15, 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.
Against this backdrop, Rubrik’s Code Guardian – an AI-powered code vulnerability scanner born out of Anthropic’s high-profile ‘Project Glasswing’ initiative – 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.
Rubrik unveiled Code Guardian at its GSI Summit in Goa last week. On the sidelines of the event, The Indian Express sat down with Anneka Gupta, chief product officer, and Alok Agrawal, chief solutions officer, to understand what it takes to build an AI security tool in such a fast-evolving field.
Here are edited excerpts from the interview:
Q: How do you build an AI security product in a field that is evolving so rapidly?
Anneka Gupta: Our product security leader did a presentation to the other Glasswing companies about what Rubrik had done and how we had used [Anthropic’s Claude] Mythos [to power Code Guardian]. After that, an Anthropic representative called me and said, “Most of the companies in Glasswing are not able to get the results out of this that you’ve been able to get.
That’s when we realised that there’s an opportunity for someone to come in and build such a product. The initial doubts that I had were: there’s already a space around code scanning, like the original SaaS vendors. But they’re not part of Project Glasswing.Story continues below this ad
In the world of AI, you have to experiment and make bets. It’s not a sure thing. At a more macro level, AI is making switching costs for software much, much lower. Until recently, we thought Mythos would be the only underlying LLM, but I think we need to adopt a multi-model approach.
Q: Code Guardian proactively scans AI-generated code for security flaws within isolated, air-gapped environments. What do you mean by air-gapped environments? How is it different from virtual machines (VMs) or micro VMs?
Gupta: Code Guardian is air-gapped from the customer. To run Mythos and scan the code, we have put in security guardrails around this environment, the sandbox in which we’re running this, how is the code getting in, what is coming out.
The value proposition is that you’re not running this 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.Story continues below this ad
Q: The OpenAI-Hugging Face attack was an inflection point. As insiders, has it changed how the cybersecurity industry thinks about AI-driven threats?
Alok Agrawal: A lot of people were predicting that something like that would happen, and the Hugging Face attack proved that the future is already here, and it showed evidence of something like that happening. It also 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.
Panel discussion at Rubrik GSI Summit 2026. (Image: Rubrik)
Gupta: The key lesson from the Hugging Face incident is that if you give models impossible tasks, they will find a way to get them done. Without the right guardrails and reinforcement during training to encourage desired behaviours and discourage harmful ones, you risk breaches that the researchers never intended. A lot comes down to the training pressures placed on the models, it’s a very complex problem.
Q: How many Indian enterprises are actively testing AI-generated code for security flaws, and how big is the preparedness gap? Story continues below this ad
Agrawal: Since Mythos is available to only 120 different companies worldwide, my belief is that the number in India is very, very small, and that is what represents a big opportunity for us with Code Guardian. Beyond tech firms, you could be a retailer with a custom app powered by custom code that will need to be scanned and secured using a tool like Code Guardian. So, we believe that this market is very big.
Q: How does Rubrik plan to price and monetise Code Guardian, and why is it relying on partnerships to take the product to market?
Agrawal: Our expertise at Rubrik is in developing software and intellectual property. We don’t have expertise in services, delivery, last-mile customisation, and that is where FDEs (forward deployed engineers) come in. That’s where our partners have a lot of strength and expertise.
Why we came up with Project Hourglass is because we want to work with specific partners that have that expertise, and we believe that this is a problem worth solving for their end clients. As part of Project Hourglass, we’ve partnered with Cognizant, Deloitte, LTM, HCL, NTT, Victron, and now we’ve expanded it to AHEAD, Phase 3, and Worldwide Technologies.Story continues below this ad
Q: What role does India play in Rubrik’s global R&D strategy, and how much is the company investing in the country?
Gupta: Around 60 per cent of our R&D is in India, and these teams are working on core products. Some of our most innovative products are being built almost entirely in India. We have very strong product management teams in Bengaluru. We co-locate product design and engineering teams to build these products, and 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.
Agrawal: From a larger perspective, India represents a huge untapped potential. We have entire product lines that are being developed and shipped from both our Bangalore and Pune offices. This is talent that is developing front-end, high-innovation products, not just back-end testing and low-level tasks.
Q: Do you worry that export restrictions and rising geopolitical tensions could affect the adoption of Code Guardian and related products?Story continues below this ad
Gupta: One of the partners I spoke to encouraged us to work with different models to give us more flexibility, given that there could be export restrictions around certain clients. We’re still early in the product development process. Our goal is to start with Mythos, but eventually make it multi-model
When building products in this environment, you have to build for flexibility and adapt to whatever comes your way.
I cannot predict what will happen, especially when it comes to government regulation. Anything can happen. So, I focus on what we can do today with the information we have and how we can ensure our approach remains flexible, without getting locked into any one way of doing things.
Q: How would you define Rubrik’s acquisition strategy, and what gaps have been filled by acquiring companies like Predibase, etc so far?Story continues below this ad
Agrawal: Every time we think that there is value to be created for a customer, we go through the Build, Acquire, Partner framework. There are parts of technology core to our offering, and so we develop it in-house organically.
Cricketer Dinesh Karthik at Rubrik GSI Summit 2026. (Image: Rubrik)
If we don’t have the expertise, we make a judgment call on how quickly we can go and acquire that expertise. It also depends on whether they have the right talent, culture, etc. With Predibase, for instance, we gained access to a pool of AI engineers, as well as its small language model (SLM). Developing our own SLM would have taken a long time.
In areas that are not core to us, where we can create greater value by working together, we partner with our alliance partners. For example, we have a strong integration with CrowdStrike, particularly its identity product which can detect malicious activity, and we can then help recover to a clean point of recovery for that particular identity provider.
The author attended the Rubrik GSI Summit 2026 in Goa, India on the invite of Rubrik.