AI Agent Assurance
Know where your customer-facing AI bot will fail, before your customers do.
TrendLab independently tests and monitors AI bots that talk to your customers: sales and support agents sold by AI startups, and shopping and service bots run by e-commerce brands. Start with a free Risk Snapshot, delivered within 48 hours of confirming scope.
What one bad answer costs
A customer-facing bot speaks for your company. When it gets something wrong, you pay.
Invented discounts
The bot offers a 40% code that does not exist, and your customer screenshots it.
Refund promises
A chatbot promises a refund your policy does not allow. A British Columbia tribunal held Air Canada to a refund policy its chatbot made up (2024).
Prompt injection
One crafted message and the bot ignores its rules.
Leaked instructions
Your system prompt, internal tools and pricing logic end up on social media.
Off-topic abuse
People use your bot as a free general-purpose AI, and you pay the model bill.
Failed security reviews
An enterprise buyer asks how you test your agent, and you have no independent evidence.
How it works
1
Free Risk Snapshot
40 to 50 outside-in tests against your public chatbot, through normal chat. Short report within 48 hours of confirming scope.
2
Agent Audit
A fixed-price deep test of security, accuracy, safety and cost, with a risk-rated findings report and a prioritized fix plan.
3
Monitoring
Every release re-tested and monthly red-team sweeps, with a scorecard and a buyer-ready evidence pack.
The Snapshot is polite and low-volume: no more than one message every 3 seconds and 60 messages in total, through your public chat window. No exploitation of your systems and no attempts to obtain real personal data.
What makes it different
Independent by design
We never test an agent we built. The test cases are hidden, so the vendor cannot see or game them.
Evidence anyone can verify
Results go into a tamper-evident, timestamped ledger. Your client can share it with buyers and auditors for independent verification.
Two independent AI judges
Responses are scored by two separate AI judges, not one, so a single model's blind spot does not set your grade.
Mapped to the frameworks buyers cite
Reports provide evidence supporting OWASP Top 10 for LLM Applications, NIST AI RMF, ISO/IEC 42001 and EU AI Act requirements. Evidence, not certification.
See what a buyer sees: the example verify page, built on TrendLab's own demo agent.
Pricing
Fixed scope, fixed price. Final scope is confirmed in a written statement of work.
Free Risk Snapshot
40 to 50 outside-in tests, short report in 48 hours.
Free
Agent Audits
Essentials
$8,000
1 agent or workflow, 2 weeks. Prompt injection and data-leak testing, 50-case accuracy test set, findings report and readout call.
Most popular
Standard
$20,000
Up to 3 agents, 3 weeks. Full adversarial suite, 200+ case domain eval suite, permissions review, cost benchmark, one re-test.
Enterprise
From $35,000
Multi-agent systems, 4 to 6 weeks. Custom suite, architecture review, framework mapping, re-test and written summary letter.
Monitoring Retainer
From $5,000 / month
Up to 2 production agents ($1,500 per month for each additional agent), or from $8,500 per month for up to 5 agents with a priority SLA. Release regression tests, monthly red-team sweeps, executive report and review call, quarterly buyer-ready evidence pack. 3-month initial term.
AI Assurance Program
From $66,000 year one
A baseline audit plus 12 months of monitoring, with re-testing and an annual re-assessment. Built for teams that need independent evidence from day one.
Prices in USD. Evaluations measure and reduce risk; they do not guarantee an AI system will never fail.
AI agent assurance questions answered
Straight answers. If yours is not here, the discovery call is free.
What is the free Risk Snapshot?
A set of 40 to 50 outside-in tests run against your public chatbot through its normal chat window, with a short report within 48 hours of confirming scope. It is polite and low-volume: no more than one message every 3 seconds and no more than 60 messages in total. We do not exploit your systems or try to obtain real personal data.
What do you need from me to start?
The public URL of your bot, your company name and an email address. You confirm you are authorized to request testing of that bot. We do not need credentials or access to your systems for the Snapshot.
What do you test for?
Invented discounts and refund promises, wrong prices, prompt injection, leaked instructions, off-topic abuse that runs up your model costs, and the weaknesses that enterprise security reviews ask about.
What does independent mean here?
We never test an agent we built for you. The test cases are hidden, so the vendor cannot see or game them. Results are recorded in a tamper-evident, timestamped ledger. The evidence can be shared with buyers and auditors for independent verification, and two independent AI judges score the responses.
Is this a certification?
No. TrendLab provides independent testing and monitoring within a stated scope. Reports provide evidence supporting your work toward OWASP Top 10 for LLM Applications, NIST AI RMF, ISO/IEC 42001 and EU AI Act requirements. They are not a certification, a compliance determination or a guarantee that an agent will never fail.
How much does it cost?
The Risk Snapshot is free. Agent Audits are $8,000, $20,000 or from $35,000 at a fixed price. Monitoring starts at $5,000 per month, and the AI Assurance Program (an audit plus 12 months of monitoring) starts at $66,000.
Who is it for?
AI startups selling agents to businesses that need independent evidence to pass enterprise security reviews, and e-commerce or direct-to-consumer brands running AI shopping or support bots. We do not test healthcare, banking or payment-data systems.
What does a report include?
Scope, sample sizes, methodology version, findings by risk level, limitations and an as-of date. Download the sample report on this page to see one.
Get a free Risk Snapshot
Tell us where your bot lives. You get a short report within 48 hours of confirming scope, no commitment.
Would rather talk it through first?
Book a callPrefer email? inquiries@trendlabanalytics.com