Is the AI Industry Hitting an Inflection Point on Data Security?

For the better part of the last three decades, the way enterprises thought about data security was straightforward: build walls around what needs to be protected. That way, the login became the door, the firewall became the alarm system. As long as you kept any bad actors outside the perimeter, the data inside was safe enough.

That logic worked when the only actors accessing corporate data were people. However, amid the rapid adoption of AI, this approach is no longer working in 2026, and the numbers tell a story most security teams already know but have not fully absorbed.

According to a recent Arkose Labs survey of 300 enterprise leaders, 97% expect a material AI-agent-driven security or fraud incident within the next 12 months. Yet only 6% of security budgets are currently destined to addressing this specific risk. The pace of AI adoption is faster than the security around it.

It’s a simple but worrying reality. The Microsoft 2026 Data Security Index, based on responses from over 1,700 security leaders across 10 countries, found that generative AI is now involved in 32% of all data security incidents. 

This week, NTT Research, a Silicon Valley-based startup from one of Japan’s largest tech companies, launched a product that is making a different kind of bet. 

SaltGrain, a zero-trust data security suite, was unveiled as the first commercial release out of the company’s newly formalized startup incubator, Scale Academy. Whether the broader industry is paying attention may matter more than the product itself.

Protecting the contents, not the building

SaltGrain’s pitch is conceptually simple. Most current security models protect the perimeter, the building that contains the data. SaltGrain protects the contents. The encryption travels with the data, meaning that a stolen file is rendered useless even if it leaves the building. No password, no usable file. The protection is a property of the data, not the environment around it.

The technology underneath is something cryptographers have been studying for over two decades: Attribute-based encryption (ABE). It was first proposed in a 2004 academic paper co-authored by Dr. Brent Waters, who now directs NTT Research’s Cryptography and Information Security Lab, and Dr. Amit Sahai of UCLA. 

Instead of locking a whole file behind a single key, ABE gives granular, line-by-line permissions. In practice, before ABE there were two options: giving full access to an AI agent to summarize a contract, but risking the information leak; or simply not giving access but missing out on the additional productivity. 

Now, there’s a third option. With this technology, the agent can read the clauses it needs without ever touching the parts marked confidential. The rules protect the data, travel with the file, not the network around it.

“SaltGrain is a data security suite that breaks the prevailing all-or-nothing file-access paradigm,” NTT Research President and CEO Kazu Gomi said at the launch. “Which is much needed in the environment where AI agents are everywhere and playing a wide variety of roles.”

The infrastructure question

The question this new development raises is bigger than SaltGrain. It is about where the next layer of AI security will actually be built.

Most of the industry’s investment so far has gone into the layers around the data. Identity and access management. Cloud security posture tools. Agent gateways that intercept tool invocations before execution. According to IDC, global security spending will reach $308 billion in 2026. AI Security is expected to take a large portion of this market, forecasted to expand to $133.8 billion by 2030. 

The majority of that spending is being directed toward monitoring, detection, and access control. SaltGrain is making the case that the encryption layer itself – not the layers around it – is where the durable answer lives. If access policies are bound to the data and travel with it, many of the gateway and monitoring tools become less necessary, not more. That is the structural argument. However, it carries implications for how AI infrastructure gets built over the next five years.

The competitive landscape is real. The Thales 2026 Data Threat Report found that 47% of sensitive cloud data is encrypted in 2026 — down from 51% in 2025 — suggesting that encryption coverage is actually losing ground as data volumes explode. 

CrowdStrike has moved aggressively in the past year to address AI agent security at the execution layer. Major hyperscalers are rolling out their own data protection tooling. The bets being placed across the industry are not the same, and not all of them can be right.

An inflection point incoming

In his opening remarks at the conference, Gomi framed the moment in terms few executives in the industry have yet been willing to use publicly.

“We are at an inflection point,” Gomi said. “The same wave of innovation that has given us so much is also generating problems we don’t know how to solve yet. Artificial intelligence is reshaping how we work, how we learn, how we make decisions. And it is raising, at the same time, urgent questions about trust, privacy, and governance, that technology alone cannot find the answers to.”

Whether this moment turns out to be an inflection point depends less on SaltGrain’s commercial reception than on whether the broader industry begins shifting its investment toward the encryption layer rather than the controls around it. 

There are signals it might. The Thales Data Threat Report noted that 59% of organizations are now prototyping or evaluating post-quantum cryptographic algorithms, and 61% cite “harvest now, decrypt later” attacks. This is the case where adversaries steal encrypted data today expecting to decrypt it once quantum computers exist. 

That last point matters. If enterprises increasingly accept that today’s data could be decrypted in five to ten years, the roadmap changes. Encryption that is both granular enough to govern AI agent access and quantum-resistant enough to last the next computing transition is starting to look less like a niche product and more like infrastructure.

Whether SaltGrain becomes that infrastructure or remains a technically elegant outlier is worth watching next. And here, the question is not about ABE technology. The harder question is whether organizations are ready to rethink security as a property of data rather than a property of networks. That implies not a software upgrade, but a shift in how all teams think, design, and audit. 

Article Co-Authored by Ana Sofia Herazo