Zscaler is expanding Project AI-Guardian, and this time it is bringing in the rest of the AI supply chain.
The company added a group of technology alliance partners across cloud, infrastructure, software, identity, data and AI platforms. The expanded roster includes AWS, CoreWeave, Databricks, Deep Cogito, Equinix, Glean, Google Cloud, OpenAI and Saviynt, along with Coforge and NTT DATA.
They join existing systems integration partners Cognizant, EY, HCLTech, Infosys, TCS and Wipro.
Translation: this is no longer just a consulting and implementation motion. This is Zscaler trying to wire zero trust into the messy, rapidly mutating AI ecosystem before every enterprise accidentally builds Skynet with a procurement department.
What AI-Guardian Actually Does
Project AI-Guardian is focused on connecting Zscaler's Zero Trust Exchange platform and AI Protect products with partner technologies so security and governance signals can move across AI-related activity. That means identities, applications, agents, data, cloud services and AI platforms all become part of the conversation.
Which matters, because enterprise AI is not one clean stack.
It is a Frankenstein quilt of cloud providers, model providers, SaaS platforms, internal apps, identity systems, data stores, APIs, agents and one executive who came back from a conference demanding "agentic everything" by Q3.
Zscaler is positioning the program around a pretty obvious problem: businesses are moving fast on AI, but security teams often have no unified way to see, govern or control the risk across all the pieces. Everyone wants the magic. Nobody wants the part where the magic emails customer data to a chatbot and calls it innovation.
The goal is to let Zscaler and its partners share visibility, enforce policy, and protect data across enterprise AI usage.
Among the capabilities Zscaler highlighted is an AI Access Graph, which maps connections between users, applications, agents and data. There is also AI risk modeling designed to identify exposure tied to AI systems and workflows.
Partner products can feed information into those systems and act on the resulting signals.
In other words, the platform is not just watching the AI party through the window. It is trying to map who got invited, who brought data, who touched what, and which agent is currently standing in the corner with production credentials and no supervision.
The Partner Network
This latest phase pushes Project AI-Guardian beyond the global systems integrator crowd and into a broader technology ecosystem.
That reflects where enterprise AI is actually going. Companies are not deploying AI from one vendor in one place with one security model. They are spreading AI across clouds, infrastructure providers, model platforms, business applications, identity systems, managed services and data platforms.
Basically, the enterprise AI stack looks less like a diagram and more like someone spilled a plate of spaghetti into a SOC dashboard.
Dhawal Sharma, Executive Vice President, AI Security and Strategic Initiatives at Zscaler, outlined the company's view of the market.
"AI is creating enormous opportunities for organizations, but it is also reshaping the threat across the security and governance landscape. Securing AI is an ecosystem effort. With the expansion of Project AI-Guardian through our technology alliance partners, Zscaler is helping customers extend zero trust across enterprise AI interactions so they can adopt AI faster while maintaining the visibility, control, and data protection they need to innovate securely."
— Dhawal Sharma, EVP AI Security and Strategic Initiatives, Zscaler
The joint work will focus on policy controls for AI workflows, protection against sensitive data leakage through prompts and training inputs, and broader oversight of AI assets such as unauthorized applications, APIs and cloud-hosted systems.
That is the right surface area, because AI risk is not just "someone used ChatGPT."
It is shadow AI. It is prompt leakage. It is agents with too much access. It is sensitive data showing up in places it should not. It is APIs nobody inventoried. It is cloud-hosted workloads talking to models without anyone knowing who approved it. It is identity sprawl wearing a fake mustache and calling itself innovation.
Industry Response
Several partners framed the collaboration around a shared problem: AI is moving from pilots into production, and the risk is spreading beyond models and apps into infrastructure, identity, data and business operations.
CoreWeave pointed to the infrastructure layer underneath AI workloads.
"As enterprises move AI into production, the attack surface expands at the infrastructure level, beyond just the application layer. CoreWeave's security is built from the silicon up, and working with Zscaler through Project AI-Guardian means customers will be able to enforce zero trust access controls at every layer of their AI stack, from compute to agent interaction."
— Jim Higgins, CISO, CoreWeave
That is the part people forget. AI does not run on vibes. It runs on compute, storage, networks, identities, APIs and access paths. If those layers are not governed, the model is just the shiny hood ornament on a burning car.
Databricks emphasized customer demand to connect security telemetry from different vendors into existing data environments.
"Customers consistently tell us they want to route their security data to their Databricks environment and extend their existing security vendor protections to our platform. Our partnership with Zscaler delivers on both fronts. By ingesting Zscaler logs into Databricks and collaborating on Project AI-Guardian, we are helping joint customers safely accelerate their AI initiatives without creating new security silos."
— Stephen Orban, SVP Product Partnerships & Ecosystem, Databricks
That one hits the operational nerve. Nobody wants another isolated console. Security teams already have enough dashboards to qualify as an aquarium exhibit.
OpenAI described the alliance as part of a broader effort to place controls, oversight and responsibility around the use of AI in security operations.
"As AI becomes an increasingly important tool for cybersecurity, organizations need systems that are not only capable, but secure, reliable, and aligned with the realities of enterprise risk management. Through our partnership with Zscaler and initiatives like Trusted Access for Cyber and Project AI-Guardian, we're advancing a shared commitment to deploying AI responsibly — combining frontier capabilities with rigorous safeguards, transparency, and human oversight. Together, we're helping security teams strengthen their defenses while building confidence in the safe adoption of AI across the enterprise."
— Scott Rosecrans, VP Strategic Pursuits, OpenAI
Saviynt focused on identity, which may be the most important and least glamorous part of this whole thing.
"AI security is an identity problem first. Zscaler stops threats in motion; Saviynt governs the identities behind them. Together, we give enterprises the control plane they need to adopt AI without losing visibility or governance."
— Vibhuti Sinha, Chief Product Officer, Saviynt
That line is the money. AI security is absolutely an identity problem first.
Because agents do not become dangerous because they are "smart." They become dangerous because someone gave them access to sensitive data, production systems, business workflows and then walked away like they just left a Roomba in the kitchen.
Bottom Line
Zscaler said the integrations are intended to reduce the burden on customers who otherwise have to stitch together separate AI and security tools themselves. It also gives partners a way to plug more directly into enterprise AI environments.
The bigger signal is clear: every serious vendor across cloud, cybersecurity, identity, infrastructure and data is now trying to figure out where it fits in the AI governance stack.
And they should.
Because AI is not staying in the lab. It is moving into workflows, infrastructure, applications and decisions. It is being connected to data, identities, APIs and agents. The blast radius is changing.
Project AI-Guardian is Zscaler's push to make sure enterprises can move fast on AI without turning the security team into the cleanup crew after a robot frat party.