Most organizations still evaluate the security of an AI platform the way they would evaluate a brand-new car. They verify that it has the safety features they require (compliance certifications), compare its performance and power (benchmarks), and confirm its quality and capabilities through other users and industry professionals (proven deployments and analyst validation). Once the boxes are checked, they declare it secure. On the day it ships, it probably is.
But attackers don’t operate on your deployment timeline. They operate in the months and years that follow, looking for vulnerabilities in operational processes as patches slip, dependencies drift, and yesterday’s hardened configuration quietly becomes today’s exposure. Real security is not a snapshot taken at go-live. It is determined by how well you patch, update, govern, and audit a system throughout its operational life.
Nowhere is this gap wider, or less understood, than in the AI space.
AI is changing both sides of the security equation. The same capabilities that help enterprises automate workflows and accelerate decisions also benefit attackers.
Why the stakes are rising now
As AI becomes more capable, the implications of failure are more significant. A chat bot informing a human employee that delivers incorrect data is a risk. An agent that autonomously acts based on that incorrect information or worse, a physical device like an autonomous car that acts unpredictably is dangerous or even life threatening. That’s why addressing the shifting security environment that surrounds AI workloads needs to be understood.
AI is changing both sides of the security equation. The same capabilities that help enterprises automate workflows and accelerate decisions also benefit attackers. Threat actors now use AI to find exposed systems, analyze new vulnerabilities, write exploit code, and launch convincing phishing campaigns in a fraction of the time those tasks once took.
AI gives attackers a dramatic increase in speed and efficiency, resulting in less time for defenders to respond. Teams that once had weeks to test and deploy a patch may now have just days. AI also helps attackers identify high-value systems, map how they connect, and adjust tactics when an attempt fails. Periodic reviews and scheduled maintenance windows cannot keep pace with that kind of adversary.
AI platforms also make attractive targets. They hold an organization’s most sensitive data and proprietary models, and they increasingly run agents that act across business systems. A single breach can expose intellectual property, corrupt the outputs leaders use to make decisions, or give attackers a foothold in critical workflows.
Private AI is the first step to enterprise grade protection.
Most enterprises stand up private AI because of security concerns, token economic efficiency, or both. They want their proprietary data, fine-tuned models, and agentic workloads under their control and not with a shared multi-tenant service. This can bring down the total cost of ownership while introducing a level of security by design, and it certainly helps maintain the reliability and accuracy of your data, but it’s just the beginning of what a secure AI strategy looks like.
When you run AI workloads privately, you also inherit lifecycle management responsibilities that provide control but can create vulnerabilities if not managed correctly. Patching, CVE triage, and driver updates can all become opportunities for human error to introduce risk to your data and operational resilience.
Why AI solutions age faster than you think
Traditional IT taught us to expect a certain cadence of vulnerabilities. The AI stack breaks that model, because it is deeper, more heterogeneous, and moves far faster than the systems it sits alongside.
Legacy infrastructure strategies gave security teams a stable foundation: a few standardized vendors, predictable applications, a monthly or quarterly patch rhythm, and a clear perimeter around servers, networks, and stored data.
AI upends each of those assumptions. Open-source frameworks and libraries evolve continuously. GPUs bring their own drivers and firmware. Models respond probabilistically, which makes misuse harder to spot. Data no longer sits in storage; it flows into models through prompts and retrieval pipelines, where attackers can manipulate it. Agentic AI adds software that reasons and acts across multiple systems on its own. Together, these shifts create a larger, faster-moving attack surface and new risks, such as prompt injection and model poisoning, that traditional security tools never anticipated.
Consider the complexity of a single AI application:
Each of these layers has its own release rhythm, advisories, and failure modes. Stitched together, change too quickly for quarterly patching. As configuration drift contributes as well, a carefully locked-down environment slowly loosens as models are swapped, pipelines are tuned, and temporary workarounds become permanent.
What effective lifecycle management looks like.
AI security must be designed for the full operational lifecycle, not just deployment. That starts with treating a few core capabilities as essential.
1. Automated patch and update cadence by default
Manual patching cannot keep pace with AI. Drivers, containers, and dependencies need continuous updates with safe rollback. That demands a unified AI strategy, yet many organizations still operate across conflicting platforms, tools, and security policies. If internal tools are outdated or unavailable, users may turn to third-party AI platforms and expose private data. This shadow AI can quickly expose security gaps that put data and profitability at risk.
2. Provenance, visibility, and control across the model supply chain
You should be able to identify every model, agent, and workload running in production, inspect its source code, and verify that it is approved, operating as intended, and staying within defined guardrails. Agentic AI can empower all this, as well as make decisions and act without human intervention. That power brings risk. Visibility, guardrails, and control reduce it while enabling productivity gains.
3. Continuous governance, not point-in-time compliance
Regulatory expectations are always changing. Frameworks and rules keep evolving, which means the “compliant” state you previously certified must be re-earned continuously. When governance is established as part of Day 2 operations, documenting and enforcing compliance to shifting requirements improves your risk profile as well as reduces administrative effort.
4. Root cause visibility
When something goes wrong in the data center, the priority is returning to an operating state as quickly as possible, and that starts with seeing what changed, who changed it, and what was exposed. That visibility must be designed in, not reconstructed after the fact like a scavenger hunt. The organizations that get this right stop thinking of security as a gate their AI platform passes through once and start thinking of it as a tool.
5. Buy the lifecycle, not just the deployment
The organizations with the best security policies implement them beyond just the purchasing of their AI platforms into actual management. They have changed the evaluation criteria from “Is this solution secure today?” to ask, “Who is accountable for keeping it secure two years from now, and how?”
That long-term view reframes the entire decision. A full-stack lifecycle-managed approach, where ongoing patching, updating, and governance are part of the platform, doesn’t simply make processes simpler. It can make a real difference in your balance sheet and risk profile. It also differentiates between a platform that was secure initially and a platform that stays secure.
The hardest part of AI security is not launching a secure platform. The challenge involves keeping that platform secure as threats, technologies, and regulations evolve.
Build AI security for the long run
The hardest part of AI security is not launching a secure platform. The challenge involves keeping that platform secure as threats, technologies, and regulations evolve. Private AI provides greater control and isolation, but those advantages depend on continuous patching, supply-chain visibility, governance, and accountability.
Lifecycle management must be central to the security strategy. Organizations should evaluate AI platforms by how reliably they can be updated, governed, audited, and restored over time. An AI platform is only as valuable as the trust it earns by delivering accurate results and protecting critical data.
HPE Private Cloud AI integrates security and lifecycle management to reduce operational complexity and strengthen platform resilience. Deploying AI securely is only the beginning; long-term value depends on solutions that keep the platform secure, scalable, and trusted over time. This helps organizations build an end-to-end AI strategy that protects critical data, adapts to change, and delivers lasting business value.
Why future AI data centers need to be open and heterogeneous
AI is evolving to be more than just a workload.
As initiatives transition from training to inferencing, from generative AI to agentic AI, AI will increasingly operate as a service and run across distributed environments, like the edge, cloud, and sovereign systems.
Future AI systems will also be heterogenous. Instead of being dominated by one architecture or platform, different workloads with different needs will require a diverse mix of accelerators, processors, and infrastructure.
Yet with the next era of AI comes a new set of IT requirements that we have to prepare for now.
AI will grow into a larger ecosystem. That ecosystem has to be interoperable to work seamlessly.
Reimagining a new operating model for AI
In order to work, AI will require a new operating model that features core elements:
Additionally, as AI scales and innovation broadens – just as we’re seeing with an emergence of neoclouds, colocation providers, and sovereign AI programs – AI will grow into a larger ecosystem. That ecosystem has to be interoperable to work seamlessly.
To ensure effective design and implementation of key capabilities, along with a resilient ecosystem, we have to take an open compute standards-based model approach to make the future of AI successful.
Making AI open and heterogeneous at every layer
At this year’s OCP Global Summit, HPE is demonstrating our ongoing commitment to contributing to open compute and ushering in a new era of AI.
We also believe that the future of AI will be more heterogenous and open compute will be critical to effectively integrate diverse silicon architectures to deliver purpose-built end-to-end solutions for AI.
As a leader in designing and delivering fully integrated systems, from the edge to exascale, we are bringing open, heterogenous technologies to every layer of the stack: compute, networking, and large-scale HPC and AI systems.
Open compute: Advancing modular, secure, AI-accelerated compute from the edge to cloud
To support a rapidly evolving AI landscape, including the transition to agentic AI and inference, we are providing enterprises with open standards-based modular compute that we continue to advance with our newly launched HPE ProLiant Gen13 servers.
Our new open compute, which includes the HPE ProLiant DL525, DL585a, XD245 and XD285 Gen13 servers, delivers a flexible design for agility to run various workloads, and scale as needed. HPE ProLiant servers are also deeply rooted with trust, and the latest generation builds on our leadership chip-to-cloud security to protect data and system integrity through automatic security from HPE Integrated Lights-Out 8 (iLO 8) with enhanced HPE Compute Ops Management.
Open networking: Ethernet-based scale-up, scale-out, and scale-across
Open Ethernet is fundamental to future AI data centers, and we’re an industry leader in open Ethernet-based scale-up, scale-out, and scale-across.
In addition to offering options in network operating systems, like SONiC, our switches have embraced open industry specifications and initiatives, such as Open Rack V3 (ORv3), Ultra Ethernet Consortium (UEC), Ethernet for Scale-Up Networking (ESUN) and Ultra Accelerator Link (UALink). These developments help build interoperable and standards-based networking for AI infrastructure.
We are the first OEM to extend open Ethernet in rack-scale level infrastructure with scale-up switches that we are delivering through the AMD Helios™ AI Rack by HPE systems. Enterprises are already adopting these rack-scale solutions to speed up their AI initiatives.
Open large-scale AI systems: Leadership-class supercomputing and AI infrastructure
Scientists and researchers are turning to solutions that effectively converge modeling and simulation with AI to accelerate scientific discoveries. These workloads require extreme scale with greater flexibility and density to maximize existing capacity and resources.
We’re building on our proven world-leading exascale-class supercomputing with next-generation open infrastructure, the HPE Cray Supercomputing GX5000, to deliver unprecedented performance with the industry’s highest density for large-scale HPC & AI workloads. The GX5000 offers various blades that feature the industry’s latest CPUs and accelerators to provide computing options that address specific workload needs.
The GX5000 also features HPE Slingshot networking based on Ultra Ethernet Consortium (UEC) specs, in addition to supporting third party components and switches.
Extending decades of direct liquid cooling leadership to AI
As a leader in direct liquid cooling, with more than 50 years of expertise and 300 technology patents, we are building on our innovation to support higher thermal capacity for high-powered CPUs and accelerators across our open compute and large-scale HPC & AI systems.
Latest innovation also includes the industry’s first direct liquid cooled ORV3 switch, a next-generation Open Rack capability, to handle extreme heat for modern AI data centers.
Join HPE at OCP Global Summit 2026 at booth D24
OCP will bring immense benefits to AI. It will pave the way for a new ecosystem that will help organizations reliably scale, achieve operational efficiency, and lower costs.
We are committed to making ongoing contributions to OCP and look forward to highlighting our vision and solutions for open technologies at OCP Global Summit.
I encourage you to join us at our booth D34 to learn more about our open, heterogenous solutions, attend our sessions, including our keynote at on Tuesday, October 13 at 10:15am PT.
Come speak with our team on how we can work together to contribute to OCP and power a new era of AI!
Everton Football Club has signed a multi-year agreement that will see HPE Aruba Networking, a leading provider for enterprise network infrastructure, become an Official Supplier of Everton Stadium, providing a secure and intelligent networking infrastructure solution.
The system, currently being installed at the Club’s new waterfront home, will include switches, access points, WI-FI and wired network management and security software, as well as network design, management and monitoring services.
HPE Aruba Networking is a business unit of Hewlett Packard Enterprise and a leading provider of next-generation networking solutions. The deal means the global edge-to-cloud company will provide a comprehensive networking infrastructure across the footprint of the Blues’ new 52,888-capacity home.
HPE Aruba Networking’s solutions will enable Everton to develop its connected stadium strategy and use state-of-the-art digital technologies to create an engaging fan experience for all visitors once the venue opens at the start of the 2025/26 season, inviting supporters to stay longer and enjoy the impressive facilities that the new stadium has to offer.
A leader in security-first, AI-powered networking, HPE Aruba Networking has a long history in providing key infrastructure for sports and event venues across the world.
Building on this experience, HPE Aruba Networking will deliver a comprehensive solution, enabling Everton to offer high connectivity everywhere across the stadium, new fan services and real-time offers, and increased speed of service at all digital interaction points.
Richard Kenyon, Chief Commercial and Communications Officer at Everton Football Club, said: “Working with HPE Aruba Networking will give Everton Stadium a best-in-class Wi-Fi network infrastructure and we are delighted to have them as an Official Supplier. Their expertise in networking some of the biggest venues in the UK, Europe and North America will be important as we develop the digital infrastructure and potential of Everton Stadium.
“A big positive of the move to our new home will be connectivity for supporters and the opportunities that will provide us to engage our partners in delivering the latest in digital experiences and services, both on matchdays and during other events held at the venue.
“I would like to thank our Partnerships and Technology teams as well as our colleagues at Elevate for the work they have put in over recent months to secure this exciting deal.”
Mark Weeks, director of UK & Ireland, HPE Aruba Networking, said: “We are proud to be joining the project to deliver a world-class venue and a world-class home for Everton.
“At HPE Aruba Networking, we know that technology is a key element to every fan experience, from the point when they enter the stadium, at times when they want to share their excitement with loved ones outside of the stadium, at half-time, when leveraging local services like restaurants and bars, and even after the match has ended.
“HPE Aruba Networking’s custom-designed and reliable network infrastructure enables Everton to stay connected with their fans throughout their matchday experience and beyond, helping the Club create a modern, interactive matchday experience.”
To find out more about Everton Stadium, one of Europe’s largest ongoing construction projects, visit evertonstadium.com
-ENDS-
NOTE TO EDITORS
About Everton Football Club
One of the 12 Founder Members of the Football League, Everton has spent more seasons in England’s top division than any other club (121) and been crowned league champions on nine occasions.
Founded in 1878 as St Domingo, the Club has also lifted the FA (Football Association) Cup five times and, in 1985, added the European Cup Winners’ Cup to its prestigious haul.
Throughout its 146-year history, Everton has been known as a family-oriented club based on proud traditions, affectionately referred to as ‘The People’s Club’. Those traditions have also encouraged innovation, ensuring Everton has remained one of the great pioneering clubs in the history of the game. The first club to construct a purpose-built football stadium, its home, Goodison Park, is one of the most revered stadiums in England. In the coming years, the Club will be looking to maintain those traditions in a new 52,888-capacity stadium at Bramley-Moore Dock, ensuring the Club’s roots remain in north Liverpool for generations to come.
As a pioneer, the Club is renowned for many firsts that we take for granted in today’s game. Everton was the first club to see its players wear shirts numbered 1 to 11 in a high profile fixture, the first club to go on an overseas tour, the first English club to install dugouts and undersoil heating, the first club to feature in a televised match, the first club to introduce a regular matchday programme, the first club to present its players with medals for winning the Football League championship and the first club to have a player, Dixie Dean, break the 60 goals barrier in a single league season.
About Hewlett Packard Enterprise
Hewlett Packard Enterprise (NYSE: HPE) is the global edge-to-cloud company that helps organizations accelerate outcomes by unlocking value from all of their data, everywhere. Built on decades of reimagining the future and innovating to advance the way people live and work, HPE delivers unique, open, and intelligent technology solutions as a service. With offerings spanning Cloud Services, Compute, High Performance Computing & AI, Intelligent Edge, Software, and Storage, HPE provides a consistent experience across all clouds and edges, helping customers develop new business models, engage in new ways, and increase operational performance.
For more information, visit: www.hpe.com
