Commvault, a data protection and cyber resilience organization, has made recommendations to help organizations stay resilient in the age of frontier AI.
Frontier models create new security risks while helping address them
As frontier models, hosted in the cloud, excel at identifying vulnerabilities at speed and compressing exploitation timelines, they also present exploitable threats to bad actors faster.
Palo Alto Networks research indicates that advanced frontier AI models are generating a significant amount of Common Vulnerabilities and Exposures (CVEs).
Further, attacks are becoming increasingly automated, enabling exploitation within minutes and not weeks.
“Frontier models change the economics of vulnerability discovery. AI models will reveal exploitable vulnerabilities at such a fast pace, remediation programs must evolve,” said Nick Patience, VP and AI practice lead at Futurum Group. “While a rigorous patching strategy remains critical, the key now is also making sure readiness, resilience, and clean recoveries are top priorities.”
“Organizations that embrace this four-step process will be better suited to take advantage of rapidly evolving AI models while also mitigating the risks,” Patience added.
Commvault’s four steps to success
The four steps include:
- Evaluate recovery risks: IT and security teams should assess if their current recovery posture can withstand fast-moving vulnerability discovery and exploitation cycles. They should look beyond whether backups exist and ask hard questions like: Can critical systems be restored cleanly? Are recovery environments isolated from compromised production systems? Are recovery plans mapped to key dependencies?
- Makes isolated recovery and air gapping the baseline: Organizations should assume that some vulnerabilities, software flaws, or third-party exposures may outpace normal remediation cycles. Maintain immutable, isolated copies of critical data and workloads, separated from production identity, network, and management planes. These copies help provide a clean fallback when patching or when remediation cannot keep pace. Additionally, organizations should pressure-test RTOs and RPOs against realistic attack scenarios and not just failure models. If your recovery time objective was set before autonomous exploitation was possible, it was set for a different world.
- Prioritize systems the business cannot operate without: Identify systems necessary to function as a minimum viable company, including identity platforms, billing systems, operational databases, and cloud services. Define the order in which they must be recovered. As AI becomes embedded in business operations, organizations should also assess newer dependencies such as data pipelines, model repositories, vector databases, and agentic workflows.
- Automate resilience and test continuously: Recovery plans cannot remain static documents in the Frontier AI era. Organizations should automate threat scanning, clean recovery point identification, dependency-aware restoration, and recovery orchestration, while regularly testing plans in isolated cleanroom environments before incidents occur.
“Resilience continues to be a high priority for us,” said Jayson Morgan, SVP of infrastructure at BOK Financial Corporation. “What matters isn’t simply whether backups exist, but whether we can recover cleanly, validate integrity, and resume operations fast when it matters most.”
ResOps for enhancing security posture
Resilience Operations (ResOps) is an operating model that can help build for a resilient future. ResOps utilizes continuous testing, measurable recovery readiness, clean recovery validation, and protection of both production and recovery environments.
ResOps is useful for business continuity during cyberattacks, outages, and AI-driven disruptions.
“AI models will continue to evolve that accelerate remediation timelines and require a new approach to readiness,” said Bill O’Connell, chief security officer, Commvault. “ResOps gives organizations a way to continuously validate readiness, advance clean recoveries, restore systems with confidence, and build resilience into the way they operate.”
Frontier models vs local AI infrastructure
While organizations have increased the use of frontier AI models to accelerate processes and boost efficiency, local AI infrastructure is becoming increasingly necessary for organizations that value resilience.
Marc Hammons, senior distinguished engineer at Dell Technologies, says that large-scale reasoning workloads still require frontier-scale models, but that local AI infrastructure can help limit the number of queries sent to frontier cloud models, which can be pricey.
Local AI provides a privacy and governance advantage while working as a practical financial optimization layer.
Further, local systems operate as environments where enterprises maintain control over intellectual property, governance device security, and endpoint validation without exposing data externally, unlike frontier cloud models.
“There’s the tokenomics that we’re going to save money by using the local AI, but since that data stays local, there’s security, there’s privacy, and there’s a governance aspect to it,” Hammons told Channel Insider onsite at Dell Technologies World. “There’s a ton of benefits to bring our data down closer to where your data is.”