RSAC 2026: When the Machines Became the Attack Surface

The RSA Conference just wrapped its 35th year in San Francisco, and if last year’s show was about separating agentic AI hype from practical integration reality, this year made one thing unmistakably clear: AI agents are not just a feature your product ships anymore. They are an attack surface your security team now has to own.

More than 43,000 attendees filled Moscone Center under the theme “The Power of Community”, a theme that landed with more weight than most conference slogans do. Because what the floor, the keynotes, and the Innovation Sandbox collectively told us is that no single vendor, framework, or platform is going to solve the agentic security problem alone. It is going to take the industry building together, and fast.

Here is what stood out.

The Storyline That Ate the Show: Agentic AI Governance

At RSAC 2025, everyone was excited about AI agents. At RSAC 2026, everyone was scared of them and building products to match.

The dominant theme across the expo floor, breakout sessions, and press releases was MCP (Model Context Protocol) governance: how organizations discover, monitor, control, and audit autonomous AI agents operating inside their environments. Nearly every major vendor touched this in some form. The sheer density of agent-focused announcements from companies like Cisco, CrowdStrike, Palo Alto Networks, BeyondTrust, SentinelOne, Snyk, and Wiz signals something important: this is not a feature race. This is a race to define a new security category.

BeyondTrust put a sharp point on why the urgency is real: research from their Phantom Labs team found that the majority of enterprises are already running shadow AI agents with privileged access that their security teams cannot see or govern. You cannot secure what you cannot inventory. That is not a future problem. That is this quarter’s incident report.

The Leaders Are Still Leading, and Executing

One of the best things about RSAC is that it sorts the companies doing marketing from the companies doing work. The established platforms showed up with real product depth in 2026.

CrowdStrike

CrowdStrike launched Falcon AI Runtime Protection and Shadow AI Discovery, extending its endpoint detection reach to every surface where AI agents operate, including SaaS, browser, cloud, and beyond. The company reports detecting over 1,800 AI applications on enterprise devices already, which tells you something about the scale of the shadow AI problem. They also added Falcon Data Security, a new solution that discovers, classifies, and stops data theft in real time across agentic workflows. And for customers who want managed coverage, Agentic MDR brings Falcon Complete’s analysts into the loop to build and deploy intelligent agents that automate high-friction security workflows.

Perhaps most notable for the broader market: CrowdStrike announced that Falcon Next-Gen SIEM now ingests and correlates Microsoft Defender for Endpoint telemetry, as Security Boulevard reports. That is a meaningful bridge for the massive installed base of Microsoft endpoint customers who want modern security operations without ripping out their existing tooling. Windows News AI has a full breakdown of the platform expansion.

Palo Alto Networks

Palo Alto released Prisma AIRS 3.0, which now covers the full agentic AI lifecycle from development through runtime. They also launched Cortex Cloud, consolidating Prisma Cloud and Cortex CDR into a unified platform, and Prisma Browser for Business, a secure browser purpose-built for SMBs. SDxCentral covered the full slate. The pattern here is consolidation with depth: Palo Alto is not adding products, they are collapsing complexity.

Google + Wiz: The Biggest Deal in Cybersecurity History

If there was one announcement that reshaped the competitive landscape before the show even opened, it was Google finalizing its $32 billion acquisition of Wiz, the largest acquisition of a venture-backed startup in history. Wiz joins Google Cloud while keeping its brand across all cloud environments, and the integration roadmap is ambitious: Wiz capabilities are being woven into Google Threat Intelligence, Google Security Operations, and Mandiant’s incident response services.

At RSAC, Google framed the combined entity as an agentic AI defense platform designed to operate at machine speed, as detailed on the Google Cloud Blog. Mandiant’s M-Trends 2026 report, also dropped at the conference, added context for why machine speed matters: cybercriminals have collapsed the defender window from hours to just 22 seconds. That is not a typo. Human response times do not win that race. SiliconAngle has more on the combined security strategy.

Microsoft

Microsoft launched the Zero Trust for AI (ZT4AI) framework alongside Agent 365, which goes GA on May 1. Their keynote centered on observability, governance, and security as non-negotiable foundations for the agentic era, and the Agent 365 launch is the execution of that thesis. Shadow AI detection and AI security dashboards round out the suite. Microsoft is positioning itself as the governance layer for enterprises already running Microsoft infrastructure, which is most of them. The AppSec Santa RSAC 2026 newsletter has a thorough breakdown of the full announcement stack.

SentinelOne

SentinelOne announced Prompt AI Agent Security, a real-time discovery and governance control plane for AI agents and agentic workflows. The platform monitors, controls, and enforces policy on agent interactions at machine speed. They paired that with Prompt AI Red Teaming for security and product teams to test first-party AI applications, and announced general availability of their Auto Investigation feature. Business Wire has the full press release, and Security Boulevard breaks down each product announcement in detail. SentinelOne continues to prove that they are not just an endpoint company. They are building a full AI security platform.

New Directions Worth Watching

Not every story at RSAC 2026 came from the incumbents.

Cisco unveiled DefenseClaw, an automated open-source agent security framework built on Nvidia’s OpenShell that scans agent skills, sandboxes code, verifies MCP servers, and inventories AI assets. They paired it with AI Defense: Explorer Edition, a self-service developer tool with red teaming for agentic workflows. Cisco also shipped an Agent Runtime SDK for AWS, Google Cloud, and Microsoft. That breadth signals a platform play, not a point product.

Snyk launched Agent Security with MCP server governance capabilities, including Agent Scan for discovering and securing MCP servers, red teaming via CLI, and Snyk Studio integration, now working with Claude Code, Cursor, and Devin. With over 300 enterprises reportedly deployed on Snyk Studio already, this is not vaporware.

Rubrik introduced the Semantic AI Governance Engine (SAGE), bringing natural language semantic policy interpretation to data governance. The concept is compelling: instead of writing rigid rules, you define policy intent and let the engine interpret it. Adaptive policy improvement over time is baked in.

Databricks made a surprise entrance into security with Lakewatch, an AI-powered SIEM built on top of their data lakehouse platform. It is a meaningful bet that the future of security operations is built on the same infrastructure organizations already use for everything else, not a separate silo.

1Password launched its Unified Access platform, designed specifically to let organizations deploy AI agents and automated workflows securely. Discovery, audit, and access control for the agentic stack. The identity layer for AI is becoming a real product category.

Constellation Research and the AppSec Santa RSAC newsletter both have strong roundups of these newer entrants and their positioning.

Innovation Sandbox: The Next Generation of the Industry

The RSAC Innovation Sandbox has launched careers and companies for two decades. Past finalists include Imperva, SentinelOne, Axonius, and Wiz, which collectively tells you how good the signal is. This year, every single finalist was integrating AI into their core product. Every single one.

The winner was Geordie AI, a London-based startup founded in 2025 by veterans of Snyk, Veracode, and Darktrace, and backed by Ten Eleven Ventures and General Catalyst. Geordie built a security and governance platform purpose-built for AI agents, giving enterprises real-time visibility into their agentic footprint, posture observation, behavior monitoring, and native risk mitigation. It is exactly the problem the rest of the show was wrestling with, solved as a dedicated product. The judges rewarded focus. PR Newswire has the full winner announcement.

Other finalists worth keeping an eye on include Token Security (non-human identity), ZeroPath (AI-powered code security), Crash Override (critical infrastructure), and Fig Security (SaaS identity governance). Hundreds of submissions narrowed to ten, as PR Newswire’s finalists announcement details. The themes they represent, namely identity, agentic risk, and AI-native security, are where the next generation of the industry is being built. Dark Reading covers the AI dominance across the sandbox well.

---

The Thread That Runs Through All of It

Zoom out from the individual announcements and a single thread ties everything together: the attack surface has changed faster than the governance model.

Security teams built their tooling, policies, and workflows around human actors operating at human speed. AI agents are non-human; they operate autonomously, they have privileged access, and they run at machine speed. The M-Trends 2026 finding about a 22-second defender window is not just a threat intelligence data point. It is an architectural requirement. If you cannot respond at machine speed, your security model has a structural gap.

The identity layer is where this gets especially complicated. Traditional IAM was built for people with credentials. Non-human identities, including AI agents, service accounts, and automated workflows, operate differently, scale differently, and represent a category that most enterprises are only beginning to inventory, let alone govern. The sessions and products at RSAC 2026 addressing non-human identity governance were among the most crowded, and for good reason. Efficiently Connected’s RSAC roundup covers the identity and post-quantum threads in depth.

Post-quantum cryptography remained a consistent thread as well, with the Cryptographers’ Panel, featuring Whitfield Diffie, Cynthia Dwork, and Adi Shamir, anchoring the academic side of the keynote program. The industry is not running out of problems to solve. Xalient’s floor report captures the grounded, community-first energy the conference carried this year.

Synqly on the Main Floor

For Synqly, this RSAC was a milestone. After spending prior years in the early-stage expo, we moved to the main conference floor for the first time, reflecting the growth we have seen and the customers we have had the privilege of supporting.

We showcased two things that got a lot of attention: our unified API technology and Synqly MCP, which brings the power of the Model Context Protocol to complex, real-world security integrations. The response was strong, but what stood out even more than the demos was the consistency of the conversations happening around the booth.

Security vendors at every stage, from fast-growing startups to established platforms, are sitting on integration debt. The problem is not that they do not know what integrations their customers need. They know exactly what they need. The problem is that they are trying to satisfy that demand while simultaneously pushing their core product forward, and those two priorities are in constant competition for the same engineering hours.

The engineers we spoke with were even more direct: they do not want to build integrations. They did not get into security to write and maintain connector code. They want to build product. Every sprint cycle that goes toward a new SIEM connector or a ticketing system adapter is one that does not go toward the thing their company actually differentiates itself on.

That frustration is not new. But it was louder this year, and the reason is everything described above. When every major vendor at the show announces new agent security platforms, new SIEMs, new governance engines, and new data pipelines, the integration surface expands accordingly. The debt compounds faster than any single engineering team can pay it down.

That is the problem Synqly exists to solve, and RSAC 2026 made the case for it in every conversation we had.

What It Means for Security Vendors

The product announcements at RSAC 2026 were impressive in depth and in range. But here is the integration reality behind all of it: every new product, every new agent security platform, every new SIEM, every new governance engine needs to connect to the rest of the stack.

The agentic era does not simplify the integration problem. It compounds it. More tools, more telemetry sources, more policy enforcement points, and a buying committee that increasingly demands pre-built connectivity before a deal closes. The vendors who figure out how to connect easily will earn more trust and accelerate faster. The vendors who require months of custom integration work on every new connector will feel the friction in every RFP.

That was the quiet theme at RSAC 2026: the operational reality of deploying these products across a heterogeneous enterprise environment is hard. The best technology in the world still has to plug in.

Synqly helps security vendors integrate faster. If building and maintaining connectors to SIEMs, EDR platforms, ticketing systems, and data lakes is slowing your team down, let’s talk.