Veeam pushes cyber resilience as AI raises data risks

As AI agents expand the enterprise attack surface, cybersecurity teams are putting greater emphasis on an AI resilience strategy built around the ability to recover quickly when prevention fails. That shift is also driving organizations to rethink data governance and security tool sprawl as AI adoption accelerates.

Cyber threats have brought greater attention to resilience in recent years, while AI is further expanding the conversation around what resilience actually means. In security, the concept is shifting away from the idea that organizations can prevent every incident and toward preparing for the reality that some attacks will get through, according to Dave Russell (pictured), senior vice president and head of strategy at Veeam Software Holding Inc.

“It means ‘OK, try to stop everything bad’, of course — but if that doesn’t work, and there’s a good probability at some point, because of the complexity of systems, because of the sophistication of attackers, and in some cases just because of accidents, there are going to be issues that arise,” Russell said. “Resilience isn’t this sort of mindset of make sure it never goes bad, but what do you do if it does?”

Russell spoke with Krista Case at Black Hat USA, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed building an AI resilience strategy through tested recovery capabilities and stronger data governance as enterprises deploy AI agents. (* Disclosure below.)

Building an AI resilience strategy

Testing is critical, because without it, organizations are relying on hope and expectation rather than knowing with confidence that they can recover. Tabletop exercises can help, but they are not enough on their own to prove that recovery will actually work, according to Russell.

“We talk ourselves into maybe a higher level of availability and resiliency than may actually be the case,” Russell said. “Recovering systems from a proof perspective, we offer recovery verification tools, but now we’re uniting that with security technologies as well on both ends of the spectrum so that we can have confidence, assured levels of recovery and resiliency.”

Strong data hygiene starts with trusted, reliable data and confidence in the ability to recover it when needed. With that foundation in place, organizations can better support the tools they deploy and the training they provide to administrative teams as they adopt AI, according to Russell.

“Those runbooks and those tools can be utilized for when weather events hit, when cyber events hit, or when AI internal events or external attacks could go in a bad direction for you. It’s still a business continuity and data resilience conversation,” Russell said. “There’s some uniqueness, don’t get me wrong, but a lot of those same steps, the answer isn’t, go deploy another thing or yet another specific, unique, one-off, isolated — something that’s not part of a combined platform taking a more holistic approach.”

Organizations are moving from theoretical discussions about AI to a more grounded understanding of how data quality affects outcomes. Redundant, obsolete or inaccurate data can lead AI agents to the wrong conclusions, with even a small amount of bad model data potentially throwing off the broader results, according to Russell.

“I think one year from now, what are we going to learn? That little things matter. Garbage collecting on the front end of production is no longer going to be a luxury item, but probably a mandate, exposure mandate, an AI quality mandate, an attack vector reduction mandate,” Russell said. “I hope we’re going to get a little better at data management and a lot better at understanding we need to prepare ahead of time.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of Black Hat USA:

(* Disclosure: TheCUBE is a paid media partner for Black Hat USA. Sponsors of theCUBE’s event coverage do not have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE