Asymptote Labs, Neo, Onyx Security and Pillar Security

About Neo

Neo sells security for AI and agentic software across every endpoint, SaaS app, and cloud service. Neo sees every AI agent, app, extension, skill, and MCP server on devices, in browsers, and across SaaS apps and cloud, checks its configuration, and enforces policy before any action runs. Neo maps an organization's teams, people, endpoints, agents, MCP servers, skills, and everything they reach.^[2][4][5]

Neo addresses a situation in which few teams can say which agents run on their fleet, what each one can reach, or what it did yesterday. Neo describes agents as acting with the employee's own access, so that most tools see the employee. Neo describes code, keys, and customer data reaching AI through prompts, pastes, uploads, and agent reads, with some tools training on what they receive. Neo describes MCP servers, skills, subagents, plugins, and extensions as arriving faster than anyone can review them, with already approved apps shipping new AI features without a review.^[5]

Neo Node enforces policy on the laptop, on macOS, Windows, and Linux. Neo Node deploys through the MDM a customer already runs, with no kernel module and no reboot, and decides each agent action with allow, ask, or block. Neo works as a layer inside the tool, in user land with no kernel component, where Neo sees the agent's actions and decides before they run. Inside Claude Code, Neo Node reads the model, MCP servers, skills and plugins that Claude Code loads. Neo Node reads the whole session rather than only the command. On managed endpoints, the hooks of Neo keep running if switched off and come back if deleted. Neo enforces before the action runs, at the tool call in tools with a Neo connector and at launch for the rest.^[5][6]

Neo Console starts from a customer's organization and its teams, taken from the identity provider. Neo Console finds settings that give an agent more reach than its work needs. Neo inventories every AI app, agent, and MCP server on each device, scopes what each can touch, and decides each action before it runs. Neo inventories every AI tool and model on devices, browsers, and SaaS, sanctioned or not, with owner, version, and publisher. Neo checks each tool's setup and version and flags drift from a customer's baseline. Neo decides every prompt, tool call, file access, and network request on the device before it runs.^[4][5][6]

Neoverse is the intelligence layer for AI apps and tools, and Neoverse is publicly available. Neoverse checks every extension, package, skill, and MCP server before it runs, and Neo blocks a hijacked or malicious release fleet-wide. Neoverse rates an unsigned server loaded by an agent as Unsafe on first sight. Neoverse rates as Malicious an MCP server that copies session memory to an outside store its listing never names, and Neo blocks it by policy when the agent loads it. Neo inspects data on the device before it reaches an AI tool, stops it or points the person to the approved tool, and flags tools set to train on a customer's data.^[5][6]

Neo covers ChatGPT, Claude, Gemini, Copilot, Salesforce, and every other AI app that teams sign into, with work and personal accounts kept separate. Amazon Bedrock, Cursor Cloud, GitHub Actions, and n8n sit under the same policy as the laptop. A single killswitch in Neo ends every active AI agent session in the estate, and no new session can start until the switch is lifted. Neo can end every session, running or new, the moment several agents break from normal behavior at once. Neo follows the agent inside each tool on the endpoint and into the SaaS apps and cloud deployments it acts through, with one product and one policy for all AI security and control.^[1][4][5][6]

About Onyx Security

Onyx Security sells a secure control plane for AI agents and models. Onyx gives security, governance, and AI teams control to protect, optimize, and measure AI agent adoption across the enterprise. Onyx was founded around a single goal: to control the most advanced forms of AI. Onyx describes its secure AI control plane as a single platform that serves security, governance, and infrastructure teams as they oversee how AI agents are used, ensure they are secure and compliant, and measure their effectiveness.^[1][2]

Onyx says enterprises were deploying autonomous agents faster than they could see, govern, or control them. Onyx offers one console for every AI agent and model across SaaS, cloud, endpoint, and code. Onyx discovers sanctioned and shadow AI across the enterprise, with visibility into every prompt, response, and agent action. Onyx identifies every agent, MCP server, AI tool, copilot, and embedded AI surface continuously, across SaaS, cloud, endpoints, and code. Onyx combines directory scanning, native platform APIs, and behavioral signature learning for browser-based AI to keep the inventory current. Onyx records every prompt, model response, and tool call by name, argument, and result, with the reasoning context preserved across turns. Onyx correlates agent activity to the invoking user and the agent owner, and captures the identity used on each tool and MCP call. Onyx forwards alerts and their session records to Splunk and CrowdStrike NG-SIEM.^{[1][2][4][5][8]}

Onyx provides the runtime and posture layer of the Onyx Secure AI Control Plane. Onyx inspects every prompt, tool call, and model response inline across browser AI, coding assistants, desktop AI agents, cloud workloads, and MCP servers. Onyx applies five enforcement actions: alert, block, mask, steer, or ask (human in-the-loop). Onyx evaluates every agent on deployment and continuously across its lifecycle. Onyx red teaming continuously attacks deployed agents, generating attack plans across reconnaissance, jailbreaking, and weaponization. Onyx identifies supply chain risks in agents, MCP servers, models, and AI assets.^[4][6]

Onyx gives an agent its own identity, separate from the invoker, when multiple users delegate to the same agent across tools. Onyx lets teams sanction approved tools and control MCP server access. Onyx offers pre-populated policy templates aligned to industry regulations. Onyx helps teams meet security standards from MITRE, NIST, OWASP, and more.^[4][7]

Onyx centralizes all AI traffic on a fully-managed, OSS AI gateway. Onyx routes across popular LLMs to automate failover, minimize latency, and balance load by task complexity and cost constraints. Onyx proxies all MCP traffic with an inline gateway, logs every request and response, and enforces guardrails. Onyx evaluates every tool call an agent makes to an MCP server against scope, identity, data class, and policy before the call lands.^[4][8]

Onyx lets teams define AI adoption targets by department, use case, or tool category. Onyx shows total AI spend, avoidable costs, and optimization opportunities. Onyx correlates AI usage, cost, risk, and outcomes to the same activity record. The Onyx AI Gateway captures per-request cost metadata at runtime: cost, tokens, model, and identity context. The Onyx Guardian Agent is a supervisory AI that continuously works across the platform to identify risks and remediate issues.^[2][4][9]

Onyx offers cloud, hybrid, or self-hosted options for sensitive data residency requirements. Onyx supports deployment with AWS VPC, Bedrock Gateway, and custom proxy configurations. Onyx offers 100+ pre-built integrations that connect to agent platforms, frontier model providers, SIEM, and cloud platforms. Onyx pulls identity context from Okta and Microsoft Entra, and endpoint management context from Intune, Jamf, and Tanium. Onyx integrates with Claude Enterprise through the Compliance API, pulling conversation content, uploaded files, project data, and activity logs into the Onyx control plane. Onyx integrates with the OpenAI Compliance API, ingesting ChatGPT Enterprise workspace activity. Onyx integrates with Amazon Bedrock Agents through Bedrock Agent configuration APIs, agent invocation logs delivered to S3, and CloudTrail data events. Onyx integrates with Microsoft Copilot Studio through the Microsoft Graph API. Onyx Security is SOC 2 Type II and ISO 27001 certified.^{[1][4][6][7][8][10][11][12][13]}

About Pillar Security

Pillar Security sells one platform to discover, test, secure and govern every AI agent across an organization. Pillar catalogs AI agents, models, prompts, frameworks, tools, MCP servers, skills and coding agents through agentless integrations. The catalog covers shadow AI and unapproved systems deployed without oversight. Pillar maps instructions, data sources, tool permissions, and external connections for each agent. Pillar traces dependencies across models and agent harness and scans agent configs, permissions, and identities.^[1][2][3][4]

For discovery, Pillar works through agentless integrations with source code repos, data platforms, and endpoints. Pillar assesses risks in AI stacks from vulnerabilities in models, frameworks, and dependencies. Pillar flags violations when unsanctioned AI appears in an environment. Pillar policy rules enforce approved models, frameworks, and data sovereignty standards across an AI stack. Pillar generates audit reports for GDPR, SOC 2, EU AI Act, and ISO 42001.^[4]

For testing, Pillar executes multi-turn adversarial attacks against complete AI systems to test tool orchestration, permission escalation, and agentic workflows. Pillar maps an AI attack surface as a graph showing relationships between agents, tools, permissions, and data sources. RedGraph maps how tools, prompts, and data combine to create exploitable surfaces specific to an environment. Pillar tests AI platforms, third-party agents, and production systems that a customer does not own or control through multi-step black box adversarial attacks against prompts, tools, and workflows. Pillar documents exploitability with attack transcripts, recordings and reproduction steps, and feeds validated findings into adaptive guardrails.^[2][3][5]

For runtime protection, Pillar analyzes user and agent intent to detect malicious behavior before it impacts production. Pillar identifies PII, PHI, secrets, and credentials flowing through AI interactions in real time. Pillar applies configurable policies to mask, block, or log sensitive data before exposure to AI models or external systems. Pillar guardrails calibrate using red teaming findings from specific applications to tune protection against the manipulation patterns detected during testing. Pillar logs prompts, responses, tool invocations, and results with timestamps. Pillar streams telemetry to security platforms for centralized monitoring.^[6]

For developer machines, Pillar deploys through an MDM and inventories the AI coding agents across a fleet: which agents run where, their versions, sandbox state and risk. Pillar runs full configuration scans for Claude Code, Cursor, Codex CLI, Gemini CLI, OpenCode and Pi Coding Agent, and detects other agents such as Amazon Q, Antigravity, Codeium and OpenClaw by fingerprint. Runtime Guardian sits in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor, and decides on every action before it runs. Pillar inventories local and remote servers across a fleet and scores each against a catalog of 23,000+ pre-analyzed servers. Pillar rolls out through Jamf Pro, Kandji, JumpCloud or any MDM that runs scripts, on macOS, Linux and Windows. On endpoints, Pillar analyzes permissions, credentials, access scopes, and auto-run settings, then provides step-by-step remediation. At runtime, Pillar monitors every prompt, tool call, and command to detect prompt injection, tool poisoning, and data exfiltration.^[7][8]

For governance, Pillar translates governance policies into technical controls that enforce approved models, licensing standards, and data sovereignty requirements at every stage. Pillar extends governance controls to external AI providers, SaaS platforms, and vendor integrations. Pillar maps governance findings, testing results, and runtime logs to regulatory requirements automatically. Pillar adds a security enforcement layer on AI gateway frameworks without re-architecting a stack. Pillar inspects every prompt and response in real time for sensitive data exposure, prompt injection, toxic content, and policy violations. For tool integrations, Pillar validates at runtime that tool calls match their declared schemas and flags anomalous behavior, like a tool accessing resources outside its expected scope.^[8][9]

Pillar can be deployed in its managed cloud, in a customer's own cloud account or data center, or as a mix of both. Self-hosted deployments include runtime guardrails, AI discovery, red teaming and agentic endpoint security. Adaptive guardrails are available in the managed cloud only today. Telemetry back to Pillar is off unless the customer enables it. A Fortune 300 manufacturer runs Pillar fully self-hosted. Pillar cites SOC 2 Type II and ISO/IEC 27001:2022.^{[1][6][7][8][10]}