For enterprises deploying AI agents
Amsterdam · Delaware
Rev · 0426—A

Underwriting the
AI agent economy.

We take on the risk, so enterprises deploy faster. Audit your AI agents with 500+ adversarial tests. Get a quantified risk score. Insure what matters.
Request a demo → Read the 2026 AI Risk Report
Based on MITRE ATLAS · OWASP LLM Top 10 · EU AI Act framework
Live sample · customer-support-agent-v4 Continuously tested · 512 scenarios
300500650750850
—
—

AI agents are creating risks
nobody can insure.

Autonomous AI agents send emails, call APIs, modify data, move money. They're non-deterministic by design. And when they fail, traditional insurance doesn't cover it.
MAR · 2026
Alibaba ROME
Rogue autonomous agent
AI agent designed for routine infrastructure tasks began mining cryptocurrency without authorization and disabled firewall protections to continue operating undetected.
Alibaba Cloud incident report
SEP · 2025
Two Sigma
$170M client losses
AI model manipulation at the quantitative hedge fund resulted in $170M of client losses before the behavior was identified and corrected.
Reuters
FEB · 2026
Lobstar Wilde
$441K accidental transfer
AI agent with tool access accidentally transferred $441K in cryptocurrency to an unintended recipient due to context confusion.
Public incident disclosure
88%
of enterprises report confirmed or suspected AI security incidents
Gravitee · 2026
21%
have visibility into what their agents can access or do
Gravitee · 2026
+$670K
extra cost per shadow AI breach vs. standard security incidents
IBM Cost of Data Breach Report

Traditional insurance wasn't built
for autonomous AI.

Cyber and E&O policies contain silent AI cover — it's unclear whether AI-specific incidents are even covered. Insurers are moving in the opposite direction: Verisk is actively filing AI exclusions from commercial general liability policies. The gap isn't closing. It's widening.
Traditional Cyber / E&O Certius Labs
Questionnaire-based assessment 500+ adversarial attack scenarios
"Silent AI cover" — unclear scope Explicit AI agent coverage
One-time underwriting Continuous risk monitoring
AI exclusions being added Coverage expands with your agents
Priced on self-reported data Priced on empirical test results
Weeks-to-months claims process Score-linked automatic triggers
Most insurance policies weren't designed for generative AI or AI agents. Karthik Ramakrishnan · CEO, Armilla AI

Audit. Score. Insure.

One platform. One workflow. From unknown risk to insured coverage in days, not months.
01 ●
Audit
Stress-test your AI agents.
Connect via API or SDK. We run 500+ adversarial attack scenarios based on MITRE ATLAS and OWASP LLM Top 10 — prompt injection, jailbreaks, data exfiltration, tool misuse, multi-agent cascade failures. We don't hand you a questionnaire. We try to break your agent.
Learn about the audit →
02 ●
Score
Get a quantified risk score.
Numerical risk score from 300 to 850, modeled on BitSight's approach to cyber risk. Six weighted categories: Security, Reliability, Permissions, Data Privacy, Compliance, Accountability. Continuously updated. Benchmarked against industry peers.
See the methodology →
03 ●
Insure
Coverage priced on real results.
Insurance premiums tied directly to your risk score. Higher score = lower premium. Built-in economic incentive to improve. Coverage up to $25M per organization. AI model errors, agent failures, data leakage, regulatory violations, IP infringement.
Learn about coverage →
The Flywheel
Every audit enriches our risk database → better scoring → better pricing → more customers → more data. The more AI agents we audit, the more accurately we can insure them.
Agent Risk Report ● Live preview 2026-04-19 · 14:32 UTC
Acme Corp · customer-support-agent-v4
Tested against 512 scenarios · last run 3m ago
300 · Critical 500 · Poor 650 · Fair 750 · Good 850 · Prime
—
—
▲ 34 pts since last week
Category BreakdownScore / 850
Security
712
Reliability
664
Permissions
527
Data Privacy
690
Compliance
374
Accountability
739
Premium quoteQuoted at current score
$184K / yr
$10M limit
▼ 31% vs. industry avg
Carrier · Lloyd's syndicate via KASKO
MITRE ATLAS
OWASP LLM Top 10
NIST AI RMF
EU AI Act · Art. 6
ISO/IEC 42001
SOC 2 Type II (in progress)

August 2, 2026 —
the AI compliance cliff.

Time until EU AI Act full enforcement
—
Days
—
Hours
—
Minutes
—
Seconds
Target · 02 Aug 2026 · 00:00 UTC
EU AI Act penalties
€35M
fixed penalty
or
7%
of global revenue
Whichever is greater. 65,000+ high-risk AI systems will need audit and certification — and Notified Bodies are still being designated.
€35M or 7%
maximum EU AI Act penalty per violation
65,000+
high-risk AI systems requiring certification
90%+
of businesses want AI insurance
Geneva Association · 2025
Singapore · Jan 2026
Agentic AI framework
First national framework for AI agents.
Colorado · Feb 2026
First US state AI law
Effective now. Algorithmic discrimination, annual impact assessments.
EU AI Act · Aug 2026
Full enforcement
€35M / 7% revenue penalties activate for high-risk systems.
See the full regulatory landscape →

Three sides of the agent economy.
One platform.

For Enterprises
Ship AI agents without waiting 12 months for underwriting.
You're deploying AI agents into customer support, code generation, financial decisions, or data analysis. Your board wants to know the risk. Your compliance officer wants documentation. Your CFO wants insurance. Certius Labs gives you all three in one workflow.
Request a demo →
For Insurance Carriers
Finally, data to underwrite AI risk.
You want to offer AI liability coverage, but lack historical loss data, assessment methodology, and ongoing monitoring tools. Our risk score is benchmarked, continuously updated, and based on empirical testing — not questionnaires. Use our data for your own underwriting, or partner with us on coverage.
Partner with us →
For AI Vendors
Close enterprise deals with a risk certification.
Your enterprise prospects are asking hard questions about AI safety. "Certified by Certius Labs" — with a published risk score — turns a long procurement conversation into a short one. Certified vendors appear in our directory for insurance carriers and enterprise buyers.
Get certified →

Built by people who break
systems for a living.

Red teaming isn't a new skill for our team — it's what we've done for 15 years. Our founder built and exited Social Links, a global OSINT platform used by government agencies and Fortune 500 companies in 80+ countries. Security assessment and adversarial testing were the core of that business. AI agents are just the newest type of system to break, understand, and quantify.
01 ● Empirical, not actuarial
We measure what happens, not what you report.
We don't estimate risk from self-reported questionnaires. We run 500+ attack scenarios against your actual agent and score what happens.
02 ● Continuous, not one-time
Your score moves with your agents.
Your AI agents change constantly — new models, new prompts, new tools. Your risk score updates automatically. Your insurance adjusts with it.
03 ● Standards-based
The frameworks the industry already trusts.
Built on MITRE ATLAS, OWASP LLM Top 10, NIST AI RMF, ISO 42001. We don't invent criteria. We apply the ones insurers, regulators, and CISOs already recognize.
04 ● EU-first, global-ready
Amsterdam. Brussels-adjacent. Lloyd's-linked.
Based in Amsterdam, next to the AI Office in Brussels and Lloyd's in London. Built for EU AI Act compliance from day one, with US expansion underway.

Deploying AI agents?
Let's talk before something breaks.

15-minute call. We'll show you how we audit AI agents, share the risk scoring methodology, and answer your questions about coverage. No sales pitch.
Request a demo → Download the 2026 AI Risk Report ↓
Email required · 40-page analysis of AI agent incidents, regulatory landscape, and insurance gaps · Free
Amsterdam-based ● Incorporated in Delaware, USA ● Backed by insurance industry advisors