Black Hat 2026: Every Major AI Security Product Launch Analyzed

At Black Hat USA 2026, the cybersecurity industry crossed a threshold. The conference was no longer about whether AI would change security — it was about how to secure the AI agents already running inside enterprise networks. Over 48 hours, more than 15 vendors launched specialized products dedicated to agent infrastructure security, marking the crystallization of an entirely new market category . Here is every major launch and the themes that tied them together.

Confirmed Product Launches

Rubrik — Agent Identity

Announced August 4, 2026. An AI-based solution to manage and control AI agents' access and permissions in real time, addressing the key obstacle of agent authorization in enterprise environments . The product monitors every agent and Model Context Protocol (MCP) at runtime using AI, and delivers just-in-time permissions per tool call. It also includes "Agent Rewind" to undo harmful actions .

Cyera — Agent Guardian

Launched at Black Hat. Monitors every tool call, database query, and reasoning step an AI agent makes — "beyond the prompt and the response" — to make agents as visible and accountable as human employees . Cyera also introduced Cyera Endpoint, bringing AI guardrails directly to employee devices for local agents operating outside corporate network controls. The product organizes its response around four stages: discovery, governance, protection, and validation . Non-human identities in Fortune 500 companies surged 480% in the prior six months, underscoring the gap Agent Guardian aims to fill .

SailPoint — Agentic Fabric (SAF)

Announced GA ahead of Black Hat. Delivers unified discovery, governance, and protection for AI agents and non-human identities (NHIs) across enterprise, endpoint, and browser surfaces . Supports 32+ agent frameworks from day one, making agents first-class governable objects . At Black Hat, SailPoint also unveiled its broader SailPoint Identity Security platform, combining human and machine identity protection in a continuous loop . Its Endpoint Agent Security and Browser Agent Security sensors expose hidden AI agents, credentials, and MCP servers .

Palo Alto Networks — PAN-OS 12.2 "Ceres"

A major firewall OS update adding more than 55 innovations aimed at attacks accelerated by frontier AI . Centers on Advanced Virtual Patching (uses AI to identify unknown vulnerabilities and deliver network protections in hours, before a traditional software patch is available), Advanced IP Defense, and six AI-powered Network Security Agents available through Strata Cloud Manager that automate onboarding, configuration, threat assessment, and troubleshooting . At the conference, Palo Alto also demonstrated NOVA (Network and Open-Source Vulnerability Analyzer), an autonomous multi-model AI harness capable of auditing codebases, writing proofs of concept, and validating severe security flaws at machine speed .

Check Point — AI Network Firewall

Introduced just ahead of Black Hat as part of firewall software release R82.20 . Extends AI security controls directly through existing physical and virtual firewalls, interpreting and securing AI-related traffic (prompts, agent behavior, sensitive business context) without requiring new infrastructure . Covers three domains: secure employee AI use (discover sanctioned and shadow AI tools, inspect prompts and file uploads in real time), oversight of Model Context Protocol communications, and protection of AI applications and LLMs including inline protection against prompt injection .

Cato Networks — Agentic Threat Prevention

Launched August 3, 2026 . Uses autonomous AI agents to predict an attacker's likely movement paths through a customer's environment, then automatically tailors and enforces protections before an AI-assisted attack can advance . Built on Cato's cloud-native SASE platform, the capability combines network and security telemetry into a single view of each customer's environment, letting the platform's AI agents anticipate how an attacker might move and adjust protections continuously across every Cato point of presence without service chaining .

SentinelOne — Purple AI Updates

At Black Hat 2026, SentinelOne showcased updates to its Purple AI platform, including Prompt AI Agent Security for protecting agentic workflows, Prompt AI Red Teaming, and the GA launch of autonomous investigations via Purple AI Auto Investigation . The company positioned itself as a unified, AI-native cybersecurity platform.

Additional Launches

Several other vendors announced products at Black Hat 2026, though verified details on Xage Security's expanded Critical Asset Protection platform, Driven Tech's ARMOR, Arctic Wolf's Cyber Resilience offering, and Snyk's autonomous penetration testing were not available within the search budget for this article. Additional notable launches included Acalvio's Deception Guardrails for AI agents (honeytokens, decoy tools, and fake infrastructure to detect prompt injection and jailbreak attempts), Sweet Security's Agentic AI Blocking, Varonis' Agent Intent-Based Access Control (IBAC), Cycode's Agentic Workflows for CVE response, and ArmorCode's expanded Agentic Control Plane with four new Anya AI agents .

Five Conference-Wide Themes

1. Agent Infrastructure Security Crystallized as a Market Category

More than 15 vendors launched specialized products for agent discovery, identity governance, behavior monitoring, and deception-based detection in just 48 hours . The sheer density of launches — from Rubrik and Cyera to Acalvio and Sweet Security — signaled that "agent security" is no longer a feature add-on but a distinct purchasing category.

2. Identity-Centric Security for Non-Human Entities

Rubrik, SailPoint, and Cyera all focused on treating AI agents as identities that must be discovered, authorized, and audited. This identity-centric approach was the dominant architectural framework of the conference. SailPoint's integration with the Claude Compliance API and its acquisition of Entro Security for NHI and credentials security further reinforced this trend .

3. Predictive, Proactive Defense

Cato's Agentic Threat Prevention and Palo Alto's AI-powered patching both aim to anticipate attacks rather than react to them. Cato's approach builds customer-specific attack models by correlating network and security telemetry, then enforces preventive controls globally . Cato also demonstrated full agentic CVE mitigation, reducing time-to-protect to 45 minutes .

4. Existing Infrastructure as the Delivery Mechanism

Check Point and Palo Alto both emphasized embedding AI security into firewalls customers already run, avoiding forklift upgrades. Check Point's AI Network Firewall requires only an OS upgrade to R82.20, with no new hardware or virtual appliances needed . Palo Alto's 55+ innovations in PAN-OS 12.2 Ceres ship as a firewall OS update .

5. AI-Versus-AI Escalation

Multiple vendors demonstrated that frontier AI models can now autonomously audit code, generate exploits, and accelerate attack timelines, compressing traditional patch windows from 50 days to near-zero . Palo Alto's NOVA demonstration was the most dramatic example: an autonomous multi-model AI harness auditing codebases and validating severe flaws at machine speed . This new reality is the urgent driver behind every product launched at the conference.