Black Hat 2026: Israeli cyber companies unveil AI defenses for the age of autonomous attacks
From protecting AI agents to predicting threats and automating responses, Israeli security firms are building new tools for a world where attackers and defenders increasingly rely on artificial intelligence.
Black Hat 2026 has become a showcase for the cybersecurity industry's response to one of its biggest challenges: artificial intelligence is making attackers faster, more adaptive and increasingly autonomous.
Israeli cybersecurity companies used the annual conference to unveil new technologies built around a common idea, traditional security tools designed around human-speed attacks are no longer enough. From protecting AI agents themselves to creating autonomous systems that can investigate threats and deploy defenses, companies are betting that the next generation of cybersecurity will rely on AI defending against AI.
Upwind targets the hidden risks inside AI agents
Upwind Security placed one of the newest challenges created by enterprise AI adoption at the center of its Black Hat announcement: securing AI agents themselves.
As companies increasingly deploy AI agents that can access code repositories, databases, cloud infrastructure and sensitive information, the instructions controlling those agents have become a new attack surface, the company said.
Upwind introduced its AI Agent Context Scanner, designed to analyze the components that determine how agents behave, including system prompts, user prompts, tools, Model Context Protocol connectors and skills.
The company argues that attackers do not necessarily need to exploit software vulnerabilities directly. Instead, they can manipulate instructions that an AI agent trusts, causing it to perform actions using legitimate permissions.
The scanner monitors instructions across developer endpoints, cloud workloads and third-party AI services, analyzing them for risks including prompt injection, compromised tools and credential exposure.
Cato Networks bets on predictive cybersecurity
Cato Networks introduced Agentic Threat Prevention, a platform designed to move cybersecurity from detection and response toward prediction.
The company said AI-powered attackers can now discover vulnerabilities, build exploit chains and launch campaigns at speeds that exceed the ability of traditional security teams to respond.
Cato's system combines networking and security data with customer-specific activity and threat intelligence to model possible attack paths and automatically create protections tailored to each organization's environment.
The company said those protections are enforced across its global cloud infrastructure without requiring additional security appliances or service chains.
Vega turns security expertise into an AI standard
Vega announced Detection Skills, an open framework designed to capture how experienced security professionals detect, investigate and respond to threats.
For decades, security teams have relied on detection rules that identify known patterns of malicious activity. Vega argues that approach is becoming insufficient as AI enables attackers to rapidly create new techniques.
Detection Skills combines detection, triage, investigation and optimization into reusable AI-powered workflows. The company is making the framework publicly available through GitHub and a dedicated website, allowing organizations and security researchers to build and share their own capabilities.
Vega said the goal is to create a common standard for AI-powered security investigations, similar to how Sigma helped standardize traditional detection rules.
Terra Security wants to close the gap between finding vulnerabilities and fixing them
Terra Security introduced Prevention, a capability designed to address a longstanding cybersecurity problem: companies often know they have vulnerabilities but struggle to fix them quickly.
Traditional penetration testing reports typically end with general recommendations, leaving security teams to determine which controls should be changed and whether those changes would actually block an attack.
Terra's AI agents test whether existing protections stop a specific exploit. If they do not, the platform generates a targeted web application firewall or firewall rule, validates it against the attack and allows security teams to deploy it with human approval.
The company said the goal is to shorten the time between identifying an exploitable vulnerability and implementing effective protection.
MIND Security brings AI agents to data protection
MIND Security unveiled AI DLP Agents, designed to automate tasks traditionally handled manually by data protection teams.
The company's platform can create sensitive data classifiers, recommend access policies, investigate data loss incidents and suggest remediation actions.
MIND also introduced a Model Context Protocol interface that allows security teams to interact with the platform using natural-language commands.
The launch comes as companies increasingly rely on generative AI tools that create, process and share unprecedented amounts of sensitive information, increasing pressure on security teams responsible for preventing data loss.
