Black Hat USA 2026: Vendors Announce Security Products and Research
Summary
Black Hat USA 2026 announcements include 1Password research finding 54% of 6,000+ AI-generated patches failed to remediate their target, Synack’s four NatJack NAT attack techniques, and a disrupted malicious skills campaign with over 1.7 million installs.
Key points
- 1Password’s Off-By-1 Labs found that 54% of more than 6,000 analyzed AI-generated patches did not remediate the target vulnerability. The company also launched a privileged-access product that creates task-scoped accounts on request and deletes them afterward.
- Synack researcher Malcolm Stagg detailed NatJack, four techniques targeting trust assumptions in NAT: hijacking active TCP connections, poisoning DNS responses, identifying ports assigned to other connections, and exhausting NAT tables to cause denial of service. The article names CVE-2026-56181 for Windows Hyper-V NAT and CVE-2026-63913 for Linux netfilter conntrack.
- CyberProof described an investigation of a browser-downloaded installer disguised as a free utility. The attack relies on user interaction rather than exploiting a software vulnerability, and uses convincing landing pages, valid code-signing certificates, and server-side control.
- Zenity and Vercel disrupted an active malicious skills campaign distributed through skills.sh; the affected skill family had more than 1.7 million aggregate installs. Zenity says its AI Total service investigates skills by executing them in a contained environment and analyzing runtime behavior.
- NeuralTrust announced an AI-agent runtime security mesh that inspects agent traffic, checks tool-authentication policies, flags prompt injections and exposed credentials, and can send alerts and logs to an enterprise SIEM.
- Cyble’s Titan update adds silicon-rooted attestation and automated attack reconstruction. RapidFort Runtime adds continuous CVE monitoring and detection of unauthorized or unexpected changes in production software.
- Other announcements included Cogent Security’s claimed frontier model VR-1, KnowBe4’s enhanced real-time coaching, Optiv’s agentic security operations service, and Vectra AI Pro for correlating security signals for SOC use of AI.
Article Details
- Event Type
- Cybersecurity vendor announcements and security research disclosed at Black Hat USA 2026.
- Impact
- The announcements covered new security products and services, research into AI-generated patches, and an investigation of the NatJack attack class affecting NAT implementations. Synack reported techniques that could hijack active TCP connections, poison DNS responses, identify ports assigned to other connections, or cause denial of service by exhausting a NAT table. The article also reported an active malicious-skills campaign distributed through Vercel’s skills.sh, with more than 1.7 million aggregate installs; Zenity and Vercel disrupted the campaign. 1Password’s research found that 54% of more than 6,000 analyzed AI-generated patches failed to remediate the target vulnerability, introduced a new one, or both.
CVE
CVE-2026-56181Microsoft · Windows 11 24h2; Microsoft · Windows 11 25h2; +2 moreTwo CVEs have been assigned to date: CVE-2026-56181, affecting Microsoft Windows NAT in Hyper-V, and CVE-2026-63913, affecting the Linux netfilter conntrack subsystem.CVE-2026-63913Vendor / Product unavailableTwo CVEs have been assigned to date: CVE-2026-56181, affecting Microsoft Windows NAT in Hyper-V, and CVE-2026-63913, affecting the Linux netfilter conntrack subsystem.
People
Vendors
1Password1Password research into AI-generated patchesCogent SecurityCogent Security has released what it claims to be a frontier AI model to match Anthropic’s Mythos.CyberProofCyberProof threat researchCybleCyble updates Titan platform with hardware attestation and automated responseGoogleOptiv has introduced Optiv Agentic Security Operations, a managed service model that integrates Google Security Operations and Wiz.KnowBe4KnowBe4 rolls out enhanced Real-Time Coaching capabilitiesNeuralTrustNeuralTrust introduces gateway-based runtime security for AI agentsOptivOptiv introduces Agentic Security OperationsRapidFortRapidFort launches continuous threat elimination solutionSynackSynack details NatJack attackVectra AIVectra AI launches AI ProVercelIn addition to its PleaseFix research, AI security company Zenity detailed an active malicious skills campaign distributed through Vercel’s skills.sh.WizOptiv has introduced Optiv Agentic Security Operations, a managed service model that integrates Google Security Operations and Wiz.ZenityZenity warns of malicious skills campaign and launches free service
Products
1Password Privileged AccessIn addition, 1Password launched 1Password Privileged Access, which extends its Unified Access platform into the PAM market.Agent GatewayOperating through its Agent Gateway, the platform directly inspects agentic traffic across the entire interaction lifecycle.AI TotalThe campaign, disrupted by Zenity and Vercel, was investigated using AI Total, a new free threat intelligence service that dynamically executes AI agent skills inside a contained environment and analyzes their runtimeBlazeAIThe platform combines endpoint activity with threat intelligence, behavioral analytics, and automated attack reconstruction (powered by BlazeAI) to form complete Indicators of Attack.Cyble TitanCyble announced an update to its Cyble Titan endpoint security platform that integrates silicon-rooted attestation.Google Security OperationsOptiv has introduced Optiv Agentic Security Operations, a managed service model that integrates Google Security Operations and Wiz.LinuxTesting found the underlying weakness across independently developed NAT implementations in Windows, Linux and macOS.macOSTesting found the underlying weakness across independently developed NAT implementations in Windows, Linux and macOS.Microsoft Hyper-VTwo CVEs have been assigned to date: CVE-2026-56181, affecting Microsoft Windows NAT in Hyper-V, and CVE-2026-63913, affecting the Linux netfilter conntrack subsystem.Microsoft WindowsTesting found the underlying weakness across independently developed NAT implementations in Windows, Linux and macOS.RapidFort RuntimeSoftware supply chain security firm RapidFort launched RapidFort Runtime, a real-time-based security solution that extends the company’s platform to create an end-to-end continuous threat elimination solution thatReal-Time CoachingKnowBe4 rolls out enhanced Real-Time Coaching capabilitiesskills.shIn addition to its PleaseFix research, AI security company Zenity detailed an active malicious skills campaign distributed through Vercel’s skills.sh.Unified AccessIn addition, 1Password launched 1Password Privileged Access, which extends its Unified Access platform into the PAM market.Vectra AI ProSecurity and observability firm Vectra AI has launched Vectra AI Pro, a new offering designed to help organizations adopt AI safely across the SOC through trusted signal intelligence.VR-1Cogent claims its VR-1 model correlates this data into a machine-readable representation of an enterprise’s security posture, with a security harness that keeps its reasoning grounded in the organization’s actualWiz CloudThe platform leverages agentic AI to analyze incoming alerts and surface environment-wide risks, which are then contextually enriched with cloud, identity, and exposure telemetry from Wiz Cloud, Wiz Code, and Wiz Defend.Wiz CodeThe platform leverages agentic AI to analyze incoming alerts and surface environment-wide risks, which are then contextually enriched with cloud, identity, and exposure telemetry from Wiz Cloud, Wiz Code, and Wiz Defend.Wiz DefendThe platform leverages agentic AI to analyze incoming alerts and surface environment-wide risks, which are then contextually enriched with cloud, identity, and exposure telemetry from Wiz Cloud, Wiz Code, and Wiz Defend.