Black Hat 2026: 450 Vendors, One Word — AI, and a Detection-Heavy Market

Every August, the Black Hat USA business hall functions as an informal census of where the security industry thinks the money — and the risk — is. This year's read, compiled by former Akamai CSO Andy Ellis and summarized by Bruce Schneier, is unambiguous: AI has stopped being a differentiator and become the baseline vocabulary of the show floor.

Ellis reviewed the messaging of all 450 exhibitors. Of those, 199 explicitly mentioned AI and 104 referenced agents or agentic technology, for a combined 235 — more than half — leading with AI or autonomous-agent capabilities as a core pitch. Even vendors who avoided the term outright were, in Ellis's framing, still competing in markets that AI has already transformed.

A market still built to find risk, not remove it

The more useful part of the analysis isn't the AI count — it's the functional breakdown underneath it. Ellis sorts security products into three buckets: tools that tell you how bad things are, tools that stop adversaries, and tools that prevent problems from occurring in the first place. His finding is a familiar one to anyone who has sat through a vendor-management backlog review: governance, compliance, third-party risk, CTEM, and threat intelligence tools made up the single largest category at 80 exhibitors — more than application security (67), security operations (55), or AI-specific governance and safety tooling (39).

That skew matters because it's a leading indicator of where security teams are spending, and buying more visibility doesn't reduce exposure on its own. A dashboard that quantifies how exposed an organization is to a given class of failure is only valuable if it's paired with the capacity — engineering time, red-team hours, patching cycles — to act on what it surfaces. An industry weighted toward detection and scoring, relative to prevention and remediation, tends to produce better risk reports and the same breach rates.

Where AI has already rewritten the category

Ellis singles out identity and access management, SaaS security, application security, and data protection as spaces where AI is now the default framing rather than an add-on. Identity is the sharpest example: the rise of agents acting under delegated human credentials — and increasingly under their own machine identities — has pushed IAM vendors well past the traditional scope of user provisioning and MFA. Any agent that can call APIs, touch data stores, or trigger workflows on a human's behalf is a new identity to govern, and the vendor response at Black Hat reflects that the problem arrived faster than the tooling matured to handle it.

For teams running AI agents in production, this is the practical takeaway: the identity and access boundary around an agent — what it's allowed to call, on whose authority, and how that's revoked — is now as load-bearing as the model's own guardrails. Vendor booths chasing this problem are a symptom of real, unresolved architecture questions, not proof they're solved.

Reading a trade-show floor as a threat signal

It's worth being precise about what this kind of analysis can and can't tell you. Exhibitor messaging is marketing, not efficacy data — a booth leading with "AI-powered" says nothing about whether the product reduces incidents. But aggregated across 450 vendors, it's a reasonably honest proxy for where the industry believes unsolved problems are compounding: agent identity, SaaS sprawl, and the gap between knowing your risk and closing it. Roughly half the exhibitors, per Ellis, didn't show at RSA Conference either, suggesting Black Hat is increasingly its own distinct signal of where practitioner-facing (rather than compliance-facing) security spend is heading.

The practical filter for buyers is the same one Ellis's trichotomy implies: ask which bucket a tool actually falls into. A platform that scores your AI agent's blast radius is doing something different — and less immediately protective — than one that constrains what the agent can do at runtime. Budget accordingly.