OX Security ships OX Cloud — AIDR for AI agents in the cloud

OX Security’s company newsroom and PR Newswire said it launched OX Cloud, a cloud security product built for risks from AI agents that access data, call tools, and act in production. The company says OX Cloud is available now to existing and new customers; AI-native pieces include AIDR (AI Detection and Response) and Agentic Attack Surface Management (AASM).

SAFETY desk — same-day company primary that a cloud/AppSec vendor put an AI-agent-aware cloud security product (AIDR + agentic attack-surface inventory) into available-now status as enterprises run agents with tool and data access in production clouds.

Availability, company: “OX Cloud is available now to existing and new OX Security customers.”. The release does not print a list price.

AI-native capabilities named on the same wire, still company: AIDR — AI Detection and Response — detects and governs AI activity across cloud and runtime environments in real time. AASM — Agentic Attack Surface Management — discovers AI agents, models, MCP servers, and non-human identities and maps what they can reach, exposing hidden access and risky behavior. MCP here means Model Context Protocol servers — the tool doors an agent can call. A non-human identity is a machine account, not a person.

CNAPP coverage, company: OX Cloud also provides what the company calls modern CNAPP coverage — Cloud Security Posture Management (CSPM), Kubernetes Security Posture Management (KSPM), Data Security Posture Management (DSPM), runtime vulnerability detection, cloud inventory, and graph-based attack path analysis — with reachability-based prioritization that filters out risks that cannot actually be reached. Alerts, the company says, open into evidence showing who or what acted, what they touched, and what else they could have reached. CNAPP is a cloud-native application protection platform — the usual cloud misconfiguration, vulnerability, and posture stack.

Problem framing, as OX Security tells it: cloud environments no longer just run infrastructure — they run agents. Traditional cloud security can see misconfigurations, vulnerabilities, exposed assets, and excessive permissions. It does not tell a team what an AI agent is actually doing: an agent can call a tool it should not, a model can touch data it should not, or an MCP server can operate outside its intended scope.

Named voice on the release: Neatsun Ziv, cofounder and CEO of OX Security. He says traditional cloud security was built to understand infrastructure — what is running, how it is configured, and where it is vulnerable — and that once AI can call tools, access data, and take actions, knowing what exists is no longer enough: teams need to know what it is doing. His other quotes are color only and stay in Sources.

Context only, not today’s dateline: OX previously described OX Cloud as part of its July 2026 AINAPP / prompt-to-runtime platform. AINAPP is the company’s name for an AI-native application protection platform — prompt-to-runtime meaning from the instruction an agent is given through to the action it takes in production. Today’s wire is the dated company launch that frames AIDR and AASM and says OX Cloud is available now as agentic cloud security. This filing is the 16 Sep 2026 available-now launch only.

About-box context, optional and company-attributed: OX Security calls itself the first AI-native application security platform built to secure software from prompt to production runtime. The about box names VibeSec, OX Code, OX Cloud, and OX Agentic Pentester as parts of that platform.

Plain English for the rest of the card: AIDR = watch what AI agents and models are doing in the cloud and flag when they cross intended boundaries. AASM = an inventory of agents, models, and MCP servers and what they can reach. MCP = Model Context Protocol servers that give agents tool access. CNAPP = cloud-native application protection platform, the usual cloud misconfig / vulnerability / posture stack. CSPM / KSPM / DSPM = posture management for cloud accounts, Kubernetes clusters, and data. non-human identity = a machine account, not a person. reachability = whether an attacker or an agent can actually get to a risk. GA / available now = offered to existing and new customers now, not a beta or a 2027 preview.