Tenet Security

**Product and the concrete problem.** Tenet Security addresses the security gap created when software agents move from generating suggestions to taking authorized actions. Enterprise coding, support, operations, and workflow agents can read repositories, query databases, invoke APIs, execute code, and pass information to other agents; ordinary identity and endpoint tools can confirm that the action came from a valid identity or process without understanding whether the agent's decision path was manipulated. Tenet's product is a security platform for this agentic layer. It is intended to discover agents and Model Context Protocol (MCP) connections across endpoints and cloud environments, show what they do at runtime, and block manipulation, hijacking, secret leakage, data exfiltration, and intent drift. The company positions the platform as a control layer that lets enterprises expand agent use without treating every autonomous workflow as an unmonitored privileged account. This is a concrete problem for security teams because an agent can perform a dangerous sequence through individually permitted tools, leaving no conventional malware, credential-phishing event, or obviously anomalous login to investigate.

**Core technology and how it actually works.** Tenet says its architecture observes three layers together: operating-system activity, network and API activity, and the agent's LLM reasoning and tool-use path. The company states that deployment does not require a gateway, proxy, software-development-kit integration, or application code changes, and that its sensor can cover endpoint and cloud environments. Its differentiating mechanism is patent-pending **Agent-Side Simulation**. When the platform sees a suspicious action, it predicts and simulates the agent's likely next moves before those moves execute against real infrastructure; if the projected path reaches a dangerous state, Tenet can intervene and generate a trace explaining the decision. The design is materially different from prompt filters or static allowlists: it attempts to assess the consequence of an agent's next action in context, including tool calls and data touched, rather than only matching a prompt or known signature. Tenet also describes a lifecycle spanning discovery, AI Detection and Response, real-time protection, risk management, and pre-production validation against real-world attacks, OWASP's Top 10 for Agents, and MITRE ATLAS. These are company-described capabilities; public sources do not establish independent benchmark results, false-positive rates, or the full sensor implementation.

**Market, customers, and go-to-market.** The initial buyer is an enterprise security or platform team deploying autonomous agents in software development, customer support, cloud operations, and internal business workflows. Tenet's wedge is an agent risk assessment and runtime proof of value: identify hidden agents and MCP integrations, show a security leader what those agents can actually do, then install a lightweight enforcement layer before broader production rollout. The official site gives one named reference, Elevate Textiles, and anonymized references from a Fortune 500 financial-services company and a more-than-$1 billion ARR U.S. legal institution. The legal customer is described as growing from two agents to more than twenty in six months while serving 24 million users, with Tenet reporting that it stopped real attacks during that expansion. SecurityWeek and SiliconANGLE separately report a Fortune 1000 deployment in which Tenet identified a runaway agent that generated tens of thousands of dollars in unnecessary token consumption over a weekend. These accounts are meaningful early proof points but remain largely customer-anonymized and company-reported. The seed capital is earmarked for product development, Threat Labs, North American go-to-market expansion, and coverage of more agent frameworks, suggesting a direct enterprise-sales motion rather than a self-serve developer product.

**Traction, funding, and third-party validation.** Tenet emerged from stealth on June 17, 2026 with a $6 million Seed round led by The Westly Group and MizMaa Ventures. The Westly Group's prior SentinelOne investment is a relevant security-market connection; CTech also reports participation from Tomer Schwartz, founder of Dazz, and Lior Tal, former CEO of Coralogix. Tenet's launch material says the product was already protecting agents in enterprise environments, with 24 million-plus end users protected and millions of agentic sessions analyzed; those figures are company claims and should be revalidated in diligence. The strongest public research signal is the company's Agentjacking disclosure: Tenet describes malicious instructions injected into Sentry error events that were returned through an MCP integration and interpreted by coding agents as trusted remediation guidance. Its controlled testing reported more than 100 agents executing the injected behavior and 2,388 organizations with injectable Sentry DSNs. The Cloud Security Alliance published a research note summarizing the attack path, an 85% exploitation success rate across tested agents, and the way the activity bypassed EDR, IAM, VPN, and web-application controls because the agent used authorized operations. That is third-party documentation of the reported research, not a full independent product evaluation. Tenet also says it open-sourced an agent-jackstop hardening repository and received 2026 industry awards, useful ecosystem signals but not substitutes for customer retention or revenue evidence.

**Founders and team background.** Tenet was founded by Barak Sternberg, CEO, and Nevo Poran, CTO. SecurityWeek, the company's announcement, and multiple public profiles say the two previously built Cisco's AI Defense research team and are alumni of Israel's Unit 8200. Before Tenet, public launch coverage identifies them with Wild Pointer, a cybersecurity company that reached seven-figure annual revenue and served Fortune 500 customers; the precise corporate history and exit terms are not publicly established in the sources reviewed. Their background is unusually relevant to the product thesis: both worked on offensive security research, presented work at DEF CON and Black Hat, and describe having seen AI systems fail under adversarial pressure before enterprise agent deployment became mainstream. Tenet says their prior work included disclosure of a zero-day in Anthropic's Claude Desktop and that the team received recognition including the Israel Defense Prize, but the exact award scope and recipient list should be confirmed directly. Public company material names additional team members, including Ron Bobrov and Shahar Marad, but does not disclose headcount. The combination of offensive research, AI-security product development, and Unit 8200 provenance is a strong technical signal; it is not evidence by itself that Tenet can build a durable enterprise platform or navigate procurement outside Israel.

**Competitive dynamics and edge.** Tenet competes in a rapidly forming market rather than an empty category. Cisco's AI Defense, Palo Alto Networks' Prisma AIRS, Protect AI, HiddenLayer, Lakera, Prompt Security, and large cloud providers all address portions of AI discovery, model security, prompt or input filtering, agent governance, or runtime protection. Tenet's claimed edge is the point of enforcement and the decision model: it aims to understand the agent's behavior across OS, network, API, and reasoning layers, then simulate the next steps and stop a harmful path without forcing traffic through a proxy. Its threat-research program can create high-value attack cases, such as Agentjacking, that expose blind spots in older controls and provide a narrative for urgent enterprise adoption. The counterargument is serious. Cisco, Palo Alto, and cloud platforms already own security telemetry, identity context, and distribution into the same buyers; they can bundle agent controls into products customers already operate. Startup competitors may also reach the same outcome through deterministic policy, sandboxing, identity, or AI security posture management rather than Tenet's simulation approach. Tenet must prove that its predictive layer works on proprietary customer agents, stays low-latency, avoids unacceptable false positives, and remains resilient when adversaries deliberately probe its decision boundary.

**Defense, security, and resilience dual-use relevance.** Tenet's core technology credibly serves both commercial enterprise and defense or critical-infrastructure contexts, so dual_use is set to true, but the public evidence is for cyber-defense relevance rather than a fielded military program. Defense organizations, intelligence environments, utilities, hospitals, and defense-industrial suppliers are all likely to deploy privileged agents that summarize sensitive data, operate infrastructure, coordinate workflows, or interact with tools across trust boundaries. A runtime layer that can identify an agent's actual behavior, constrain tool calls, and stop an authorized-but-manipulated sequence could improve resilience against prompt injection, malicious telemetry, poisoned documents, insider-assisted automation, and agent-to-agent escalation. The Agentjacking research is particularly relevant because it shows how an external data source can become an instruction channel without conventional code execution or stolen credentials. Tenet's Israeli origin and Unit 8200 background create credible access to allied cyber-defense conversations, but no public source reviewed here establishes an IDF contract, government deployment, classified accreditation, FedRAMP authorization, or critical-infrastructure customer. Defense diligence would additionally require evidence of operation in disconnected or classified environments, tamper resistance, deterministic human-approval boundaries, data residency, supply-chain security, and export-control posture.

**Growth stage, trajectory, and key diligence risks.** Tenet is classified as **early**: it is a newly public Seed company with promising production references and a technically focused founding team, but no disclosed revenue, customer count, renewal rate, valuation, headcount, or independently audited deployment metrics. The trajectory is attractive if enterprises treat autonomous agents as a new privileged computing layer and need an enforcement product that existing EDR, IAM, WAF, and prompt-guardrail systems cannot provide. Key diligence points are: (1) validate the anonymized customer claims, paid status, agent counts, session volumes, and blocked-attack evidence; (2) test whether Agent-Side Simulation adds measurable protection beyond sandboxing, least privilege, deterministic policy, and standard runtime telemetry; (3) quantify latency, operating cost, false positives, and failure behavior when the platform itself is degraded; (4) determine how much proprietary data or intellectual property is defensible beyond a patent-pending architecture and threat research; (5) assess competitive bundling from Cisco, Palo Alto, Microsoft, AWS, and hyperscaler model platforms; (6) verify the legal entity, U.S. headquarters, Israeli operating footprint, and cross-border data handling; and (7) examine whether a small team can support high-trust enterprise and government deployments. Tenet merits high-priority monitoring because the problem is strategically important and the founders have unusual evidence access, while its commercial durability and defense readiness remain unproven.