Organizations are increasingly deploying AI agents, giving users tools that can supercharge productivity but also open up a new cyber-risk front. As
Lavi Lazarovitz, vice president of cyber research for security vendor
CyberArk,
stated, “the more autonomous and interconnected these AI agents become, the larger the attack surface they create.”
“Unlike traditional automation or bots that follow a rigid script, AI agents can make decisions, learn from their environment, and act autonomously to complete complex tasks,” Lazarovitz wrote. “Their adoption is accelerating. By 2027, multi-agent environments are expected to be the norm, with the number of agentic systems doubling in just three years.”
A growing number of security vendors are rolling out products and services that enterprise security teams and MSSPs can use to help organizations track their AI agents, from the data they are accessing to what they are doing with it. Such
visibility into agentic AI operations is important, according to
Yossi Altevet, co-founder and CTO for
DeepKeep, a five-year-old startup that offers an end-to-end AI security platform that the company last month made available on
Google Cloud and
Amazon Web Services (AWS) marketplaces.
“Agents are gaining more autonomy, accessing company data, interacting with each other in the ‘Internet of Agents’ and increasingly making decisions that impact the business,” Altevet told MSSP Alert. “This creates a very different risk surface compared to traditional applications, as the behavior is non-deterministic and the attack surface can be unclear.”
Enterprises today “risk losing control and visibility into what their agents are accessing, how they are behaving on a daily basis, and how each element can be manipulated,” he added. “This is where the security gap exists.”
Enter AI Agent Scanner
The Israeli company this week added AI Agent Scanner to its portfolio of AI security offerings, giving organizations and MSSPs immediate visibility into what AI agents can access and the tools and data they can interact with, and details where potential vulnerabilities are. Users get a visual risk map to understand the range of their agents’ exposure and threats, outlining how each could be leveraged by attacks and the defensive actions they need to take.
It also delivers runtime protection for agentic frameworks, identifying where users should place AI firewalls and guardrails. Among the frameworks it supports are
Microsoft-based frameworks,
Salesforce’s Agentforce,
OpenAI agents,
CrewAI,
Amazon Bedrock AgentCore, and
n8n.
Four Layers of Agent Protections
DeepKeep will also expand its agent security capabilities to cover the full AI lifecycle, including adding a red teaming solution, which, along with visibility and mapping, make up two of the four key layers to building protections around AI agents, according to Altevet. Runtime protection and identity make up the other two layers.
Red teaming allows security teams to “proactively identify vulnerabilities in how the agent behaves under different conditions and interacts with multiple data sources and AI models working simultaneously,” he said. “It is crucial that these layers are connected and working in harmony with each other, in order to keep up with the constantly evolving threat landscape.”
Such capabilities will be key as the adoption of agentic AI grows.
Gartner analysts last year predicted that by 2028, agents will make at least
15% of work decisions, and 33% of enterprise applications will include agentic AI.
MSSPs at the Intersection
AI agents – through the automated actions they can take across tools, data sources, and infrastructure – create a new class of risk that is out of reach of many existing security services, Altevet said. This is where MSSPs and MSPs will play an important role
“They already operate at the intersection of infrastructure, identity, and security, and are well positioned to extend that into the agent layer,” he said. “That includes assessing how agents are configured, what access they have, and how they interact with internal systems, as well as continuously monitoring their behavior at runtime to detect misuse, unintended actions, or policy violations.”
Operationalizing Governance
Agentic AI will require ongoing governance that goes beyond one-time validation. Security services providers will be important to operationalizing such governance, which includes maintaining visibility, enforcing policy, and securely scaling the use of such autonomous systems.
DeepKeep’s new solution will help, with the most immediate value coming from the visibility it will offer without needing a deep familiarity with each agentic AI framework, he said.
“From there, it enables them to deliver new services, such as agent risk assessments, automated red teaming, ongoing monitoring, and recommendations for where guardrails or controls should be applied,” Altevet said. “Internally, many MSSPs are also starting to use AI agents in their own operations, and the same visibility and controls apply there.”
