How EōS Fitness is balancing AI risk and innovation
Equipped with a backup environment that allows the company to store and recover critical data securely and quickly, the chain of gyms has become a first-mover in agentic AI
A gym may not be the first type of business that comes to mind when thinking about artificial intelligence. But inside EōS Fitness, a fast-growing gym chain, agentic AI is already reshaping operations, from customer service to personal training.
On the company’s website, an AI-powered chatbot helps prospective customers navigate membership options and gym information. Inside its fitness centres, trainers are using AI-driven workout tools that tailor exercise routines based on a member’s fitness level and goals. Within the business, employees are increasingly experimenting with AI to automate workflows and improve efficiency.
“We’re taking a cautiously optimistic approach,” says Aaron Gette, the company’s chief information officer. “We’re doing more and more with it and trying to figure out the best way forward—while also mitigating the risk.”
That balance between innovation and security is becoming a defining challenge for organisations deploying agentic AI. Businesses feel pressure to move quickly in order to stay competitive, but greater autonomy and access to data also introduce new operational and cyber-security risks.
For EōS Fitness, those risks became apparent early.
The new risks of the agentic era
When EōS deployed its first agentic chatbot, the company discovered that controlling AI systems was more complicated than simply defining approved data sources.
Even after teams restricted the chatbot to a designated knowledge base, the system sometimes pulled information from outside sources or generated inaccurate answers through hallucinations.
“Even when you told it, ‘Look at just this knowledge base to answer these questions,’ it still found ways to pull information from other places,” Mr Gette says.
The experience highlighted a broader reality now facing businesses: AI systems introduce entirely new security and governance challenges.
Attackers are increasingly using AI tools to automate phishing campaigns, mimic employee behaviour and scale cyber-attacks. At the same time, organisations face internal risks tied to employee use of AI tools and the integrity of enterprise data.
“The democratisation of these tools does not equal literacy,” Mr Gette says. Employees may unintentionally expose sensitive information to public models or unknowingly corrupt critical datasets while using AI-powered automation tools.
Mr Gette also points to emerging research1 showing that large language models (LLMs) can introduce errors into documents and spreadsheets as they manipulate or reassemble files. In a business environment, even small inconsistencies can cascade across finance, sales or operations.
“If somebody realises that somewhere along the way an LLM corrupted a document or dataset, now every downstream outcome may be incorrect because the original data was compromised,” he says.
That risk is particularly significant for EōS because of how heavily the business depends on cloud-based systems and shared operational data across its gyms and corporate operations.
If somebody realises that somewhere along the way an LLM corrupted a document or dataset, now every downstream outcome may be incorrect because the original data was compromised.”
Aaron Gette, chief information officer at EōS Fitness
Building cyber resilience in the cloud
Running almost entirely in cloud-based environments, EōS Fitness needed a resilience partner that could help it recover effectively from incidents across a multi-cloud infrastructure. The company turned to Rubrik.
Before the partnership, EōS had relied largely on native cloud tools and standard backup practices that were sufficient for day-to-day operations, but not designed for the growing complexity introduced by AI.
Rubrik added what Mr Gette describes as a critical missing layer: a centralised, air-gapped backup environment capable of securely storing and rapidly recovering essential business data and systems.
“It gave us another layer for offloading our infrastructure somewhere that we knew we could effectively bring back online very quickly,” he says.
The platform also became central to EōS’s broader cyber-resilience strategy. The company now conducts tabletop exercises and incident simulations designed to stress-test recovery procedures and prepare teams for real-world disruptions.
“We were able to really challenge ourselves as an organisation to be ready for an incident,” Mr Gette says. “And to be able to bring our critical infrastructure back online and keep the business running should something occur.”
That capability has become increasingly important as AI systems become embedded in day-to-day operations.
“It’s not really a matter of if, it’s when,” Mr Gette says of cyber-incidents. “When it does happen, can you effectively execute your incident response plan? Can you bring the business back online?”
Unlocking AI—with guardrails
Now fully integrated with Rubrik, EōS finds itself equipped to recover data and systems more quickly and precisely. That is critical because of how quickly a cyber-attack could otherwise put key business functions at risk.
The recovery capability also serves as a safeguard against internal human error—the lesser-discussed AI vulnerability that, Mr Gette says, is every bit as real. With enhanced reversibility from Rubrik, EōS can restore clean versions of critical documents and datasets, whether they have been accidentally altered by a team member or maliciously attacked by an outsider.
Everyone is going through the exact same challenges. How do you unlock value, how do you get the tools in the hands of people, and what do you actually do with those tools?
Jesse Ruegg, vice-president of AI operations at EōS
“Everyone is going through the exact same challenges,” says Jesse Ruegg, vice-president of AI operations at EōS. “How do you unlock value, how do you get the tools in the hands of people, and what do you actually do with those tools?”
The balance between moving fast and moving securely remains elusive, Mr Ruegg and Mr Gette agree, but the company’s investment in resilience has helped it take another step towards unlocking the benefits of AI.
References
- 1.