TL;DR: RSA Conference 2026 kicked off March 23 with one clear message: "agentic AI" is the new buzzword, and vendors are racing to sell autonomous security systems that act without human approval. Simbian unveiled the "first autonomous SecOps platform." Microsoft announced AI agents that investigate threats solo. Innovation Sandbox finalists include companies building entire AI workforces. Federal agencies are still boycotting. The sales pitch is appealing. The surveillance implications are worth examining.
The Agentic Takeover
Day 1 at Moscone Center delivered what everyone expected: 44,000+ attendees, 700 vendors, zero federal officials, and "agentic AI" repeated so often it lost all meaning by lunch.
Here's what that actually means. Traditional AI tools analyze data and give you recommendations. Agentic AI tools analyze data and take actions, without waiting for a human to approve them. They query databases, block IP addresses, shut down processes, send emails, and modify firewall rules on their own.
At RSA 2026, vendors aren't just talking about this. They're selling it.
Simbian: "Autonomous SecOps" Goes Live
The biggest announcement came from Simbian, which unveiled what it calls the "industry-first reasoning-based approach to unified offensive and defensive security."
The platform runs three AI agents:
- AI SOC Agent: Investigates alerts, correlates events, and resolves incidents (24/7, no human needed)
- AI Pentest Agent: Runs penetration tests against your own systems autonomously
- AI Threat Hunt Agent: Proactively searches for attackers hiding in your network
Simbian claims these agents "follow the same operating procedures as experienced human experts." The company says its Context Lake technology lets the agents reason across security and non-security data to "confront AI-armed cyberattacks at machine speed."
Translation: the AI watches everything across your systems and decides what's a threat without asking you first.
The SOC Agent and Pentest Agent are "generally available for immediate deployment." Simbian is at booth N6567 through March 26.
Microsoft: Security Copilot Gets Autonomous
Microsoft's Vasu Jakkal delivered a keynote titled "Ambient and Autonomous Security: Building Trust in the Agentic AI Era." The company announced expanded capabilities for Security Copilot that let AI agents investigate threats across Defender and Sentinel without waiting for human direction.
One new feature: the RSA Advisor for Admin Threats agent. It monitors admin accounts and alerts security teams about suspicious activity. The word "alerts" is doing heavy lifting here: the agent decides what's suspicious and can take automated response actions.
Microsoft framed this as "deep observability" to address "AI-powered threats." The practical reality: more AI watching employee behavior, more automated decisions about what's suspicious.
WitnessAI: Watching the AI Watchers
Here's where it gets interesting. WitnessAI is pitching a product that monitors all the other AI agents running in your company.
The company claims it can:
- Uncover "shadow AI": unapproved AI assistants employees are using
- Catalog every AI application, MCP server, and agent in your organization
- Monitor "real-time interactions" between employees and AI systems
- Enforce "guardrails" on what AI agents can and can't do
WitnessAI markets this as "security and governance." From an employee privacy perspective, it's comprehensive surveillance of how workers interact with AI tools, plus the ability to block or modify those interactions in real-time.
Their booth is demonstrating "new Agentic AI Security capabilities" all week.
Innovation Sandbox: The Startup Previews
The Innovation Sandbox contest happened this morning, with 10 startups pitching to win the "most innovative cybersecurity startup" title. Several finalists focus on AI agent security:
- Geordie AI: A governance platform for AI agents that gives enterprises "real-time understanding of their agentic footprint" and the ability to "observe agent posture and behavior." Think: tracking what every AI agent in your company is doing, all the time.
- Charm Security: An "Agentic AI Workforce" that uses behavioral psychology to prevent social engineering and fraud. The agents "guide real-time prevention, intervention, and resolution."
- Token Security: Focused on machine identity security as AI agents proliferate.
The winner was announced around noon. Regardless of who won, the finalists show where security investment is heading: AI agents monitoring AI agents, all the way down.
OpenText: "Trust by Design"
OpenText hosted a session called "Deceive, Detect, Defend" about "data-centric security" that combines data discovery, encryption, and behavioral analytics with identity monitoring.
The company introduced its "trust by design" model for securing AI agents, which includes:
- Identifying sensitive data
- Securing identities
- Monitoring behavior
- Securing apps
- Incident response
"Monitoring behavior" is the key phrase. To secure AI agents, you need to watch everything they do. To watch everything they do, you need to watch everything employees do with them.
The Surveillance Pattern
Strip away the buzzwords and RSA Day 1 reveals a clear pattern:
- AI agents are coming to corporate security: autonomous systems that act without human approval
- These agents need comprehensive visibility: access to network traffic, endpoint data, user behavior, and system logs
- Monitoring AI agents means monitoring employees: every interaction, every query, every action gets logged
- New tools are emerging to monitor the monitors: creating surveillance layers all the way down
The vendors aren't hiding this. They're selling it as a feature. "Real-time understanding of your agentic footprint." "Comprehensive monitoring on admin accounts." "Shadow AI detection." Each capability requires watching what employees do.
The Federal Silence
All of this is happening without federal participation. CISA, FBI, and NSA officials who could offer guidance on responsible AI deployment, or raise red flags about surveillance overreach, aren't here.
The agencies boycotted RSA after former CISA Director Jen Easterly was named CEO. Their absence means 44,000 security professionals are getting vendor pitches without federal counterbalance.
No government voice asking: "What happens when the autonomous SOC agent decides a legitimate employee is a threat?" No one from CISA warning about AI agents making split-second decisions that affect real people's jobs and lives.
Just vendors, buyers, and a lot of talk about "machine speed" decision-making.
What's Coming
RSA runs through March 26. Day 2 brings Microsoft's Vasu Jakkal keynote on autonomous security. Day 3 features sessions on securing AI agents across enterprise environments. The Cryptographers' Panel (Diffie, Dwork, and Shamir) will discuss encryption in an era where AI agents need to read everything to protect it.
Gartner says 40% of enterprise applications will have AI agents by year end. What happens after that? Your company's security operations run themselves. Your network traffic gets analyzed by AI that decides what's suspicious. Your interactions with AI tools get logged and monitored.
The machines are hiring themselves. RSA 2026 is where companies come to sign the contracts.
References
- Business Wire - Simbian to Unveil Industry-First Autonomous SecOps Platform at RSA Conference 2026
- Microsoft Security Blog - Your Complete Guide to Microsoft Experiences at RSAC 2026
- WitnessAI - RSA 2026
- PR Newswire - Finalists Announced for RSAC Innovation Sandbox Contest 2026
- OpenText - OpenText to Showcase Secure Agentic AI Across the Enterprise at RSAC 2026
- Cybersecurity Dive - Federal Agencies Abruptly Pull Out of RSAC
- RSA Conference 2026 Official Site