Axiomat
Software that formally verifies AI agents' answers to company data queries so businesses can trust them without manual checks.
AI-safety startups
November 2026
4
Stacktrace
Monitoring and control layer that lets teams see, bound, and intervene in what their AI agents do.
Asim
Standardized tests that show robotics companies where their robots fail under changing weight, friction and materials before deployment.
Routant
Zero-trust access controls letting outside AI agents work in enterprise customers' private systems, with approvals and audit logs.
October 2026
8
Mirrors
Scoped corrections that stop production AI agents from repeating recurring mistakes, for engineering teams running customer-facing agents.
Latent
Flags likely-wrong LLM answers for finance, legal and healthcare teams before customers see them.
Armorer Labs
Security testing and runtime monitoring for AI agents at companies in regulated industries.
Funding: Backed by Paul Vixie (amount not stated)
Sphinx
A hook that quizzes developers on AI-written code before they push to production.
Proof of Human
An API that continuously checks whether users are humans or bots and AI agents, for websites and apps.
Keythorn
Insurance for AI agents, plus stress tests and monitoring that help companies fix agents and pass procurement.
Funding: Backed by a16z speedrun and Okta co-founder Frederic Kerrest; no amount stated
Switchfrog
Software that lets companies see, govern and monetize AI agents using customer accounts on their websites.
DeepMark
Audio watermarks in AI agent voices that survive phone calls, for banks and call centers to verify agents.
September 2026
3
Traceforce
Security monitoring that shows companies which AI apps employees use on their devices and blocks risky actions.
LLM Attention Visualization
A tool that visualizes attention patterns inside large language models, for people studying how they work.
Weedout
Safari extension for Mac that hides YouTube videos labeled "Made with AI" from feeds, search and Shorts.
August 2026
1
Decawork
Small company-specific models that approve or block each action AI agents take, for businesses running agents.
July 2026
5
Formally verified 3D CSG
Mesh intersection code for 3D solid modeling, proven correct in Lean 4 so reviewers need only read the spec.
OneCLI
Open-source gateway that keeps API keys away from AI agents and enforces request policies for developers running them.
Fabraix
Automated red-teaming agents that find security holes in customer-facing AI agents for large enterprises.
Clawk
Disposable Linux VMs that let developers run coding agents away from their laptops.
Energent AI
Agent that independently re-verifies every number in AI-generated results and shows its work, for analysts and teams.
June 2026
5
Are You in the Weights?
A site that queries many AI models in parallel to show how strongly they recognize you personally.
Lattice Health
Software that monitors deployed medical imaging AI for hospitals, tracking accuracy, drift and bias against radiologist reads.
Claw Patrol
Network firewall that lets AI agents reach production systems while humans and LLMs approve destructive actions.
Wato
A shared workspace where teams give AI agents permissioned access to company tools, context, and reusable workflows.
BentoLabs AI
Monitoring for teams running production AI agents, catching silent failures and suggesting prompt or harness fixes.
May 2026
6
AGI Deploy Co
Insurance and security coverage for businesses running AI agents in production.
Forge
Open-source guardrails that make small self-hosted LLMs reliable at multi-step tool-calling workflows for local users.
Git for AI Agents
Open-source history tool for Claude Code sessions that records what the agent changed and why.
Clawvisor
Authorization layer letting teams give AI agents Gmail, Slack and Drive access without exposing credentials.
Chronicle Labs
Staging environments that replay recorded production events so enterprises can safely test changes to their AI agents.
Klaimee
Liability insurance and risk certification for companies deploying AI agents that take actions autonomously.
April 2026
2
Agent Vault
Open-source proxy that holds credentials so AI agents can call services without ever seeing the secrets.
Silmaril
Prompt-injection defense for AI agents, plugged into frameworks like LangGraph, for companies running agentic apps.
March 2026
5
Zerobox
Command-line sandbox that blocks file and network access and hides API keys from commands, for developers running AI agents locally.
OneCLI
Open-source proxy that gives AI agents placeholder keys and swaps in real credentials, so developers' agents never see secrets.
A context-aware permission guard for Claude Code
Permission hook for Claude Code developers that classifies each tool call by effect and allows, asks, or blocks it.
Agentic Fabriq
Identity, scoped permissions, approval flows and audit logs controlling what company AI agents can access and do.
Envariant
Interpretability SDK that lets foundation model builders analyze and steer their models' behavior.
February 2026
6+
enveil
Tool that hides .env secrets from AI coding assistants so developers' API keys stay out of their context.
Steerling-8B
An 8B language model that can explain any token it generates, according to its one-line listing.
Local-First Linux MicroVMs for macOS
Disposable Linux virtual machines for macOS that boot in about a second, mainly for sandboxing AI agent code execution.
TwentyTwo
Debugging assistant that pinpoints where developers' AI agents went wrong and why.
Salus
An API that checks AI agent actions at runtime, blocking wrong ones before they execute, for developers building agents.
Sentrial
Monitoring and debugging for engineering teams running AI agents in production, catching loops and hallucinations live.