The Future Arrived Faster Than
Security Was Ready For
RSAC 2026 was the 35th annual gathering of the global cybersecurity community at San Francisco’s Moscone Center — and the clearest signal yet that the industry faces a reckoning it cannot defer. Here is what happened, what was overlooked, and where SecOps is headed next.
Biggest Highlights
There was very little conversation on the floor of the Moscone Center that did not route back through AI. Every panel — whether focused on investment, products, identity, or offensive capability — returned to it. Approximately 40% of the full conference agenda was AI-weighted across every cyber domain, making it less a track and more the event itself.
AI vs. AI: The New Security Paradigm
Kevin Mandia of Ballistic Ventures argued that AI agents, already deployed in red teaming exercises, are capable of operating at scale and speed, compressing attack-response cycles from five days to five minutes. “We have to take humans out of the loop,” he said. “AI versus AI is the new paradigm.” The critical implication: this symmetry cuts both ways. The same AI capability that allows defenders to automate response at machine speed is simultaneously being weaponized by adversaries to automate attacks. Speed alone is no longer a competitive advantage; the architecture underlying the response is what will determine outcomes.
The SIEM Is Dead
Live on the RSAC floor, Databricks CEO Ali Ghodsi stated plainly: “AI will kill the SIEM in 2026.” It was one of the most widely circulated claims of the conference and one of the most consequential for SecOps teams heading into H2. For context: SIEM (Security Information and Event Management) has been the backbone of enterprise threat detection for two decades, aggregating logs and alerts across an organization’s environment to surface anomalies. The argument against it is structural. Traditional SIEMs were built for rule-based correlation of known threat patterns. AI-native platforms can reason across unstructured data, investigate in real time, and adapt without manual rule authoring, making the legacy model increasingly difficult to justify on cost, speed, or efficacy grounds.
From Copilots to Autonomous Analysts
The conversation moved decisively beyond AI copilots toward autonomous agents triaging alerts, investigating malicious activity, isolating hosts, and patching software. The honest reality on the floor was that early deployments have been uneven; many organizations are running agentic tooling in supervised modes with significant human checkpoints, and the gap between vendor demos and production behavior remains wide. The trajectory is nonetheless clear, the investment is accelerating, and the teams building governance frameworks for agentic SOC workflows now will be significantly better positioned than those waiting for the technology to mature in isolation.
Active Defense and Cloud Consolidation
Two distinct but related stories. On the keynote stage, Sandra Joyce reframed the industry’s posture from passive detection toward active defense: proactively disrupting threat actor infrastructure, imposing operational costs on adversaries, and moving security from a reactive discipline to a strategic one. Separately, Google used RSAC to frame its $32 billion Wiz acquisition as the architectural foundation of a unified, AI-ready cloud security platform. The pairing matters because active defense at scale requires the kind of cross-environment visibility that only consolidated cloud-native platforms can provide. New agentic automation was introduced in Google Security Operations to begin connecting those capabilities.
Geordie AI Named Most Innovative Startup
Geordie AI took the top prize at the Innovation Sandbox, recognized for its work in AI-powered threat detection and autonomous security response. The Innovation Sandbox is widely regarded as one of the industry’s most credible early signals for where enterprise security investment is heading, with past winners going on to define entire product categories. Also notable on the floor: Jazz, building a next-generation DLP engine purpose-built for AI environments with an agentic investigator that collects rich contextual signals to drive investigations, won CrowdStrike’s 2026 Cybersecurity Startup Accelerator. Both point to the same direction: the new frontier in security tooling is built around AI-native architectures, not AI features bolted onto legacy products.
Four NSA Directors on Offensive Cyber
One of the more substantive national security conversations of the week brought together multiple former NSA Directors and Commanders to discuss lessons from offensive cyber operations. The central thread: offense has outpaced defense not primarily because of technology gaps, but because of institutional ones. Decision cycles, attribution processes, and response authorities were all identified as areas where adversaries currently hold structural advantages. The session landed with particular weight given that U.S. federal agencies notably reduced their presence at RSAC this year, a withdrawal that many attendees read as a signal of shifting priorities at the government level precisely when public-private coordination on cyber threats is most needed.
Everyone only wants to talk about AI, and if you aren’t doing AI, investors don’t want to talk to you.
— Yoav Leitersdorf, YL Ventures · RSAC 2026 Executive DinnerWhat Was Understated
In a conference where AI consumed the oxygen in nearly every room, several critically important threads did not receive the stage time their urgency warranted. SecOps leaders should treat these as the sleeper issues of 2026.
Non-Human Identity as the New Perimeter
Quietly the most consequential undercurrent of the conference. Non-human identities (AI agents, bots, service accounts) now outnumber human identities in enterprise environments, yet most organizations still lack the visibility and controls to govern them at scale. SailPoint’s founder framed it precisely: do we even understand the intent of an AI agent, and do we have guardrail policies capable of reasoning about it? This received far less main-stage attention than the agentic AI hype it directly relates to.
Post-Quantum Cryptography Readiness
Sessions urged IT leaders to begin inventorying cryptographic assets and developing migration strategies now, with the conversation shifting decisively from “if” to “when.” Given the stakes — sensitive data encrypted today could be decrypted by future quantum capability — PQC was notably underrepresented relative to its long-term impact. The organizations that start now will have a meaningful head start by 2027.
Confidential Computing
With AI data centers projected to grow from a $25 billion industry today to as much as $360 billion by 2030, confidential computing is emerging as a critical infrastructure trend. It received almost no main-stage attention, yet underpins the trusted data foundations that every agentic AI deployment requires.
Why Phishing Still Works in 2026
In a technically saturated conference, one session drew on Nobel Prize winner Daniel Kahneman’s work to explain the persistent human vulnerability at the root of most breaches. We operate in fast, automatic, intuitive “System 1” thinking all day, making humans exploitable regardless of technical controls. In an AI-saturated event, this grounded human-factors work got lost in the noise. It shouldn’t have.
Supply Chain Security Regulation
A Cisco/Splunk panel explored how increasingly strict global regulations and rising supply chain attacks are reshaping CISO strategy, with legal, technical, and operational implications that extend well beyond the SOC. It was largely drowned out by AI conversation, but the regulatory pressure it describes will arrive regardless of the hype cycle.
Previous discussions were centered around ‘we have AI.’ RSA 2026 made it clear that the focus has matured to: ‘now let’s secure it.’
— Olha Kolomoets, VP of AI Engineering, Apriorit · RSAC 2026What SecOps Should Expect: H2 2026 → 2027
The widening gap between innovation velocity and organizational readiness was the subtext of nearly every conversation at RSAC 2026. The trends below are not abstract forecasts; they are already reshaping day-to-day security operations. The question is whether your team is ahead of them or behind them.
Legacy SIEM Faces Existential Pressure
The Databricks CEO’s claim is a leading indicator, not a fringe opinion. AI-native security data platforms are emerging as serious challengers to legacy SIEM architecture. Expect vendors to race toward LLM-powered detection and investigation pipelines, and expect procurement conversations to shift accordingly by Q4 2026.
Agentic SOC Becomes Operational, Not Theoretical
SecOps teams will need playbooks for AI agents acting as autonomous analysts, and for securing those agents against compromise and manipulation. Organizations need to evolve identity strategies to support machine and agent identities at scale, and integrate cyber resilience with broader business continuity frameworks.
AI Red Teaming Goes Commercial
The AI red teaming conversation surfaced at RSAC with more commercial urgency than expected. Expect purpose-built AI red teaming tools and services to proliferate through H2 2026, moving from research curiosity to budget line item. Teams that build internal AI red teaming capacity now will be better positioned when adversaries operationalize the same capabilities.
IAM Must Scale to Machine-Speed Governance
IAM vendors face a fundamental rebuild pressure. Traditional identity models were designed for humans at desks, not for thousands of AI agents operating autonomously at machine speed. SecOps teams should begin auditing non-human identity sprawl now. The organizations that have clean inventory going into 2027 will have significantly more control over their exposure surface.
PQC Readiness Enters the Procurement Conversation
Organizations that haven’t started a cryptographic asset inventory will face increasing regulatory and insurance pressure to do so through 2026. PQC migration planning will move from CISO talking point to concrete project budgets by 2027. The risk isn’t quantum computers today; it’s “harvest now, decrypt later” attacks that are already underway.
The Intelligence-Response Gap Is the Core Problem
The organizations that will succeed will not necessarily be the fastest adopters; they will be the most disciplined in how they bring clarity, structure, and intent to complexity. Providers and internal teams that can move beyond tool-centric services and anchor their value in operating model design, risk alignment, and cross-domain orchestration will define the competitive landscape heading into 2027.
First Principles Don’t Expire
In the rush to position for an agentic, AI-driven future, RSAC 2026 largely treated Zero Trust, Security by Design, and Least Privilege as assumed background rather than active conversation. That’s a mistake worth naming. Every headline problem on the show floor (shadow AI, agent identity sprawl, prompt injection, uncontrolled access surfaces) is at its root a first-principles failure. The Cisco team acknowledged it plainly: despite years of momentum since the Jericho Forum and Google’s BeyondCorp, many organizations still haven’t fully implemented Zero Trust. That implementation gap doesn’t close by layering agentic tooling on top of it.
Programs like Booz Allen’s Thunderdome, one of the most rigorous real-world Zero Trust implementations on record, demonstrated that these aren’t legacy frameworks waiting to be replaced. They’re the architectural bedrock that makes new capabilities trustworthy. Security by Design means AI governance gets embedded at the start of deployment, not retrofitted after a breach. Least Privilege means an AI agent gets exactly the access its task requires, nothing more, regardless of how autonomous it becomes. SecOps teams that treat these principles as settled and boring will find themselves chasing the same exposures with more expensive tools. The teams that operationalize them rigorously are the ones that will actually be ready for 2027.
Leave a Reply
You must be logged in to post a comment.