Akto Named Sample Vendor in Gartner’s Inaugural Hype Cycle for Secure AI, 2026

On October 12, 2026, Gartner published its first Hype Cycle for Secure AI. The report maps the technologies organizations need to secure the AI applications and agents they adopt and build. Akto is named a Sample Vendor in two of its innovation profiles: AI Security Testing and AI Runtime Defense.

Being recognized in both matters because the two categories address the security of the same AI applications at different stages. AI security testing finds weaknesses in AI applications and agents before they ship. AI runtime defense watches and controls them once they are live. The report links them directly, recommending that runtime defense buyers require integration with AI security testing. Akto was built to do both.

Where securing AI stands in 2026

The report describes a market where AI adoption is real, but production deployments are still early. Gartner cites its 2026 Surviving the Agentic AI Era survey: only 33% of organizations report having enterprise-built, customer-facing GenAI agents in limited or broad production. In the same survey, 50% of CISOs reported uncertainty about the long-term costs of specialized AI security controls, and 43% see those controls as too expensive for their perceived value.

Funding is a real constraint. Among organizations piloting or implementing specialized AI security technologies, almost half either have not secured funding or have had to divert budget from existing cybersecurity initiatives. Gartner reports this at 50% for offensive AI security testing and 47% for AI runtime defense.

The report also notes that AI tools change so often that projects pivot mid-course, which slows every category on the curve. We read that as a requirement. AI security controls have to keep working as the models, frameworks, and agent tooling underneath them change.

The two categories Akto is named in

The two profiles share a maturity level and market penetration but sit at very different points on the curve.

AI security testing uncovers vulnerabilities and exposures in AI-enabled systems before deployment. Gartner's definition covers offensive techniques, such as automatically generating and running adversarial prompts, and scanning AI components such as model repositories, libraries, frameworks, and notebooks.

AI runtime defense enforces intent-based policy and detects anomalies in AI applications and models while they run. It inspects content for abuse and attacks such as prompt injection, and Gartner notes that these tools are evolving to protect AI agents, including against agent tooling abuse.

Akto covers both. On the testing side, it red-teams internally built AI applications and agents before they ship. On the runtime side, it enforces guardrails, applies least privilege on each agent action, adds MCP security controls, and shows what agents do as they execute.

Why runtime defense needs testing behind it

Runtime defense sits at the Peak and testing sits earlier on the curve, but we believe the report gives three reasons to fund them together.

Start with incident risk. One of the report's strategic planning assumptions says that by 2028, organizations that skip preproduction offensive testing on GenAI deployments will face twice as many cybersecurity incidents as those that test proactively. Runtime controls catch attacks in production, but testing finds the weaknesses before attackers get the chance.

Testing also makes runtime controls better. Gartner observes that runtime defense increasingly integrates with AI security testing from the same provider, and that this integration improves runtime controls by tuning them on the results of adversarial assessments. Its user recommendations for runtime defense go further and advise buyers to require integration with AI security testing so they can evaluate runtime controls before an application or agent is deployed.

Then there are false positives. The report warns that real-time alerts on text inputs can be noisy, and that prompt injection detection will take time to reach good accuracy, much as SQL injection and XSS detection did. Testing gives security teams a known set of attacks to measure their runtime controls against, so they can see the false-positive and false-negative rates instead of guessing.

We built Akto around that loop. Red-teaming results help you define your runtime controls, and what runtime sees in production informs the next round of tests.

How to evaluate runtime defense vendors at the Peak

A category at the Peak of Inflated Expectations has more vendors, more claims, and less proof. The report is candid about this. It notes that many runtime defense providers are small and their effectiveness remains unproven, that several are being acquired, and that runtime tools have not yet caught up with the new problems autonomous agents create. Its companion report, the Hype Cycle for AI-Driven Cybersecurity, includes a profile on AI washing.

We'd welcome the scrutiny. These are the questions we'd ask any runtime defense vendor, Akto included:

1. Can you show benchmarks, test data, and test results, including false-positive and false-negative rates?

2. Does your runtime defense connect to adversarial testing, and do test results change how the controls are tuned?

3. What can the product see? Gartner notes that a control deployed in front of an application sees only user inputs and outputs, and that some applications need several entry points in the stack.

4. Does the product cover discovery and posture as well as input and output monitoring?

5. How does it handle retesting as models, prompts, and agent tools change?

Each question comes from the user recommendations and obstacles Gartner lists for the two categories.

Where security leaders can start

Based on Gartner's recommendations for both categories, we'd prioritize four steps.

1. Inventory your AI assets first. Gartner points out that testing depends on visibility into every model, including locally downloaded ones. You can't test or protect an agent you don't know exists.

2. Put AI security testing in the CI/CD pipeline. AI applications change fast enough that one-time tests go stale. Gartner recommends automated testing in CI/CD and tools that support continuous testing and flexible retesting.

3. Require testing before runtime controls go live. Use adversarial test results to evaluate and tune runtime defense before an application or agent is deployed.

4. Assign clear ownership. Gartner notes that AI applications often reach preproduction without proper security testing because several teams share responsibility. Name an owner for AI application security before the next agent ships.

Akto's Gartner recognition in 2026

This listing adds to Akto's earlier Gartner recognition this year. Akto was named a Pioneer in the 2026 Gartner Emerging Market Quadrant for AI Application Security, Startup Vendors. It was also named a Sample Vendor for AI Security Testing in the Hype Cycle for AI Services, 2026, and for AI Runtime Defense in the Hype Cycle for AI Governance Technologies, 2026, and the Hype Cycle for Data Security Technologies, 2026.

Akto is also named in the companion report published the same day, the Hype Cycle for AI-Driven Cybersecurity, 2026.

Link to the report - Hype Cycle for Secure AI, 2026

Source: Gartner, Hype Cycle for Secure AI, 2026, Jeremy D'Hoinne, Dionisio Zumerle, Jeff Arias, 12 October 2026.

Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner's research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose. GARTNER is a registered trademark and service mark, and HYPE CYCLE is a registered trademark, of Gartner, Inc. and/or its affiliates in the U.S. and internationally and are used herein with permission. All rights reserved.