Sweet Security Raises $75M for Unified AI and Cloud Defense

An Israeli startup led by the former CISO of the Israel Defense Forces raised $75 million to integrate artificial intelligence security capabilities into its CNAPP platform.

See Also: Agentic AI and the Future of Automated Threats

Tel Aviv-based Sweet Security plans to leverage its existing runtime CNAPP infrastructure, eBPF sensors and patented deviation detection engine to provide deep insights into AI agent behavior, said Eyal Fisher, co-founder and chief product officer at Sweet Security. AI agents, while non-deterministic, can still be baselined and monitored like other runtime elements, ensuring companies retain visibility and control, he said.

“We found out that there is a huge need for the market, and we found out that we are winning POCs, meaning that whenever somebody is trying us, they take us,” Fisher told Information Security Media Group. “So, now the logic is, ‘Okay, if that thing is working, let’s do more.’ We are going to triple our go-to-market in order to bring more POCs and to win more customers.”

Sweet Security, founded in 2023, employs 82 people and has raised $120 million, having last completed a $33 million Series A funding round in March 2024 led by Evolution Equity Partners. The company has been led since its inception by Dror Kashti, who spent 21 years with the Israeli Military Intelligence and nearly four years as head of the Israel Defense Forces’ Joint Cyber Defence Division (see: AI Security Goes Mainstream as Vendors Spend Heavily on M&A).

Why CNAPP and AI Security Need to Come Together

Evolution Equity also led Sweet’s Series B round since its reputation with large enterprise customers is especially important in the security space – where trust and validation matter deeply, Fisher said. He credited Evolution Equity’s day-to-day strategic guidance, ability to open doors to new customers and alignment with Sweet’s long-term vision as instrumental to the company’s success.

“It’s important that your investors have a good reputation. We have Fortune 100 customers that are into understanding who is backing you. If somebody like Evolution is backing you, it’s a hint you’re doing something right,” he said. “They have good connections and support us in accessing customers. We get guidance and help along the way. We are hyper-growing right now and need people with experience.”

Current CNAPP tools are insufficient for the new risks introduced by AI agents operating autonomously in cloud environments, Fisher said. He believes security platforms should offer both runtime protection of infrastructure and proactive control over AI behaviors. AI agents behave like code, but with non-deterministic elements, and require real-time behavioral monitoring, not just scanning or static analysis.

“We use AI in two ways,” Fisher said. “We harness AI for security, and the second is we secure AI. We do both. Our platform was built as an AI-first platform. We have an LLM in the heart of our solution. It’s the essence of what we do.”

CNAPP must evolve to integrate AI security capabilities as enterprises deploy AI agents within their cloud environments, and he said Sweet Security aims to unify cloud runtime protection and AI agent oversight. Sweet Security is extending its runtime detection and baselining capabilities to apply to AI agents, Fisher said, tracking deviations from normal behavior and enforcing runtime policies.

“What we see that is missing in AI security is first, visibility – understanding what’s happening in your environment,” Fisher said. “That’s what we get from our customers. They say the security team says, ‘We are blind. We don’t know what’s happening in our environment.’ Second, you would like to understand the behavior of the agents. Third, we are going to put guardrails on those agents.”

How Sweet’s Ability to Baseline Workloads Secures AI Agents

Sweet’s sensors sit close to the kernel, giving them deep visibility into workload behavior, helping the company assess baseline workloads quickly and detect meaningful deviations from normal behavior, Fisher said. If an AI agent normally accesses a customer database but begins accessing an unfamiliar system or sensitive directory, Fisher said Sweet can detect and flag that deviation.

“An eBPF sensor sees the kernel, sees everything that’s happening in your environment – system access, memory access, CPU usage, secrets usage and everything,” Fisher said. “It’s the same. All you need to know is what the agent is doing. The same sensor can do that as well. See which files are being accessed by the agent, what the agent is doing, how it behaves. It’s the same stuff.”

CISOs and security teams should continue to be the central buyers for Sweet Security’s offerings since splitting responsibility for security between traditional cloud teams and new AI-focused groups would lead to inefficiencies and gaps. AI use may involve data scientists, engineers and business teams, but securing AI workloads falls squarely within the domain of security architects and detection and response teams.

“Security for AI is going to still be a problem of the security team,” Fisher said. “The SOC is going to be the one to follow alerts. It should go into the same flow, the same person, the same everything. Otherwise, you’re going to have two different sub-organizations not talking to each other. It’s unrealistic.”

Cloud-native customers tend to be well-informed about cloud security but often find tools from cloud providers insufficient, and Fisher said Sweet can offer more advanced runtime protection and behavioral insights. Sweet offers traditional industries such as banking and insurance a way to modernize their security posture while maintaining the clarity and control they’re used to from on-premises environments, he said.

“One thing that CISOs are concerned about is being considered those who stop AI adoption,” Sweet said. “That’s why the CISO is in such a stress. You don’t want to be the CISO that is holding back the organization.”