RSAC 2026 Reveals an Industry Racing to Secure AI Agents It Has Barely Begun to Deploy
The cybersecurity industry's largest conference was dominated by agentic AI security products, even as data shows only 5 percent of enterprises have moved AI agents into production.
Overview
The 35th RSA Conference wrapped up in San Francisco on March 26 with a single theme dominating the show floor: how to secure autonomous AI agents before they become the enterprise attack surface of the decade. From Cisco and Palo Alto Networks to SentinelOne and a roster of ten Innovation Sandbox finalists, virtually every major vendor unveiled products designed to govern, monitor, or contain AI agents operating inside corporate networks.
The irony was hard to miss. According to Cisco, 85 percent of enterprises are experimenting with AI agents, but only 5 percent have moved them into production. The industry is building security tooling for a deployment wave that has not yet arrived at scale.
What We Know
The largest announcements centered on creating identity, access control, and runtime security for AI agents that operate with significant autonomy.
Cisco’s Three-Pillar Framework
Cisco introduced what it called a framework for securing the “agentic workforce,” built around three pillars: protecting the world from agents, protecting agents from the world, and detecting and responding to AI-related incidents at machine speed. The company released DefenseClaw, an open-source security framework that scans Model Context Protocol (MCP) servers, plugins, and other technical resources that agents use to perform tasks. According to SiliconANGLE, developers can install DefenseClaw in approximately five minutes, and it can block specific MCP servers within two seconds without restarting the affected agents.
Cisco also announced Zero Trust Access for AI agents through Duo IAM, enabling organizations to establish agent identities and enforce fine-grained, time-bound access permissions. Cisco’s chief product officer Jeetu Patel framed the stakes plainly: “The difference between delegation and trusted delegation is the difference between bankruptcy and market leadership,” according to Cisco’s newsroom.
Industry-Wide Product Wave
Cisco was far from alone. Palo Alto Networks advanced its AI security platform with Prisma AIRS 3.0, a single platform designed to manage the primary threats and risks of AI applications and autonomous agents across their entire lifecycle. SentinelOne launched Prompt AI Agent Security for real-time discovery and governance of AI agents, alongside Prompt AI Red Teaming for testing applications against prompt injection and related threats.
Datadog released its Bits AI Security Analyst for general availability, claiming to reduce security investigation times from hours to as little as 30 seconds. Stellar Cyber introduced agentic AI capabilities for its autonomous SOC platform, and Panther announced general availability of its AI SOC Platform, in which AI agents have native access to the platform’s data lake and detection engine to investigate and triage alerts autonomously.
Innovation Sandbox Goes All-In on AI
The RSAC Innovation Sandbox competition, long regarded as a barometer for where the industry is heading, drove the point home: every single finalist was integrating artificial intelligence into its product, according to Dark Reading. Among them, Charm Security is building what it calls an “Agentic AI Workforce” to prevent scams and social engineering, while ZeroPath focuses on autonomous vulnerability detection.
What We Don’t Know
The gap between experimentation and production deployment remains significant. Cisco’s own data showed that while 85 percent of enterprises are experimenting with agents, the 5 percent that have deployed them into production face governance challenges that remain largely unsolved. According to Dark Reading, nearly half of respondents in industry surveys believe agentic AI will represent the top attack vector for cybercriminals and nation-state threats by the end of 2026.
It remains unclear whether the security tooling announced this week will mature fast enough to keep pace with the deployment wave that vendors themselves are forecasting. Many of the products announced at RSAC are in early availability or preview, and several critical capabilities from Cisco’s Splunk-integrated SOC agents to SentinelOne’s red teaming platform are not scheduled for general availability until mid-2026.
Analysis
RSAC 2026 marked a notable shift in the cybersecurity industry’s relationship with AI. Previous conferences focused on using AI to enhance existing security operations. This year, the conversation pivoted to a fundamentally different question: how to secure the AI itself. The challenge is that AI agents, by design, need broad access to enterprise systems to be useful, a requirement that conflicts directly with decades of security architecture built on the principle of least privilege.
The volume of announcements suggests vendors are not waiting for the deployment wave to arrive before staking out their positions. Whether enterprise buyers will adopt agent-specific security tooling before they adopt agents themselves, or whether the two will arrive in tandem, will determine whether the industry gets ahead of this particular threat or, as has happened before, ends up playing catch-up.