Reading an AI Vendor's Security Page: What Matters

If you’re evaluating AI tools for your business, you will eventually land on a security page. These pages exist to reduce concerns, and they often do - but not always for the right reasons. Most vendor security pages are written for enterprise IT teams: dense with certifications, architecture diagrams, and compliance frameworks. If you’re a small business owner who wants one answer - whether your customer records and business data are actually safe - here is what the page can tell you, what it can’t, and what you need to ask for directly.

What a security page is actually for

A vendor’s security page is a starting point, not a verdict. It puts the vendor’s best posture forward for a specific audience: IT professionals running vendor assessments, compliance officers checking regulatory boxes, and cautious customers looking for reassurance before signing up.

For small business owners, the result is a page that answers some questions you don’t have and sidesteps others you do. What certifications does the vendor hold? On the page. Does the vendor use your data to train their AI models? Typically in the terms of service, not the security page. What happened during the vendor’s last security incident? Almost never mentioned.

Reading a security page well means knowing what it was written to communicate, and what it quietly leaves out. The page is a marketing document as much as it is a technical one. That is fine - it just means treating it as the first step in a short conversation, not the end of your research.

Which certifications mean something, and which don’t

Some terms on a security page come from third-party verification. Others are phrases the vendor chose to describe themselves.

SOC 2 Type II is an audit conducted by an independent firm covering how a vendor actually handles data across a 6- to 12-month operating period. Type II matters more than Type I: Type I only checks whether security controls exist at one point in time, while Type II checks whether they hold up consistently over time. A vendor with SOC 2 Type II can typically share the full audit report on request, under a mutual NDA. If they are reluctant to do this, that is itself worth noting.

ISO 27001 is an international standard for information security management. It requires an external audit and annual reviews. A vendor who holds it has a systematic, documented approach to security - not just a one-time configuration they made at setup.

HIPAA Business Associate Agreement (BAA) is a legal contract, not a certification. If your business handles health-adjacent data, a vendor who processes that data needs to sign a BAA before you start. A security page that claims HIPAA compliance without a clear offer to provide a BAA is missing the part that actually has legal weight.

Phrases like “enterprise-grade security,” “bank-level encryption,” and “military-grade” are different. They have no standard definition and no auditing body behind them. A vendor using them isn’t necessarily doing anything wrong - they may have genuinely strong security. But the phrases themselves confirm nothing. When you see them without corresponding certifications, you are reading the vendor’s self-description, not an external verification.

If a vendor has no third-party certification at all, that is not automatically a reason to walk away - many early-stage or niche tools haven’t completed a SOC 2 yet. What it does mean is that you are relying on their word alone, and your questions about data handling become more important.

The data policies that matter for your business

Past the certifications, three questions directly relevant to small businesses are often not answered on the security page itself.

Model training. Does the vendor use your inputs to improve or train their AI? This is one of the most common concerns for businesses passing anything sensitive through an AI tool, and it is also one of the most inconsistently handled. Many vendors do this by default at lower price tiers, with an opt-out available in enterprise agreements. If you are putting customer conversations, internal documents, or proprietary processes through the tool, knowing this before you connect production data is not optional. The post on where your data goes when you use AI tools covers how this typically works in practice.

Data retention. How long does the vendor keep your data after you stop using the service? A 30-day deletion window is different from an 18-month retention policy. If the answer is “as long as necessary for our services,” that is worth pressing on before you agree.

Data residency. Where are servers located? For businesses in regulated industries or those with customers in the EU, this has compliance implications. For others, it shapes which privacy framework technically governs your data.

None of these are typically on the security page. They live in the terms of service and privacy policy, which the security page usually links to. Follow those links and read them before you decide - or at minimum search them for the terms that matter most to your situation. For businesses in fields like healthcare, the data privacy and HIPAA considerations that apply before signing any AI vendor agreement are worth understanding before you reach the terms of service.

What you can only learn by asking

The security page describes a vendor’s posture under normal conditions. It tells you nothing about how the vendor behaves when something goes wrong - how quickly they respond, how specifically they communicate, and whether the response matches what the page promised.

A vendor who has experienced a breach and handled it well - notifying affected customers quickly, explaining exactly what was compromised, making documented changes afterward - may have a stronger actual security story than one with no reported incidents. Untested systems have unknown failure modes. The page does not tell you which kind of vendor you are dealing with.

Three direct questions are worth asking before you connect production data to any AI tool:

Has the vendor had a security incident in the past two years? If so, how did they notify customers, and how quickly?

Who is the named contact for security concerns, and what is the typical response window?

Can they share their most recent SOC 2 Type II report under a mutual NDA?

These are reasonable requests for any vendor who will handle sensitive data. The answers matter. So does how the vendor responds to being asked at all. A security team that replies quickly with specific documentation and a follow-up contact behaves differently from one that redirects you to the same page you already read. Both responses tell you something real.

The gap no page can close

There is a version of vendor evaluation that treats the security page as a checkbox: it exists, the certifications look right, approved. That version skips the part that matters most.

The difference between a vendor who passed an audit and a vendor who runs a genuine security culture shows up in the questions they were not expecting. It shows up in how a support ticket about unusual account activity gets handled, in whether the team knows what happened and can explain it, in whether they seem like they have been asked these things before. None of that is on the page.

This matters more for AI tools than for many other software categories, because what flows through them tends to be more sensitive than businesses realize at first: customer conversations, internal planning notes, financial queries, process documentation. The convenience of the tool is usually obvious. The question of whether the vendor treats your data with the same care is not.

For businesses thinking through their current AI stack or building the right vendor selection process, what a private AI setup looks like in practice is the right starting point for anything that warrants keeping data fully in-house. The post on reading AI vendor feature claims covers the marketing-claims side of the same evaluation. And the AI and data privacy overview for regulated fields addresses what the BAA requirement actually means in practice. The services overview at Elements AI has more context on how vendor and tool selection fits into a broader AI strategy for small businesses.

VK, an AWS Certified Solutions Architect and founder of Elements AI, a Castle Rock studio, offers a free 30-minute scoping call for businesses working through these decisions. The starting point is usually not which tool to pick - it is which questions to ask before the contract is signed.

Frequently asked questions

What does SOC 2 Type II certification actually mean for a small business?

SOC 2 Type II is a third-party audit of a vendor’s security controls over a 6- to 12-month period. It means an independent auditor verified how the vendor actually handles data, not just how they say they do. Type II is more meaningful than Type I, which only checks whether controls exist at one point in time, not whether they hold up consistently.

Can an AI vendor use my business data to train their models?

Often yes, by default. Most enterprise AI tools include a data processing agreement with an opt-out, but smaller subscription tiers sometimes do not. This is typically in the terms of service or privacy policy, not the security page. If you cannot find it clearly stated, ask before connecting any production data.

What does enterprise-grade security mean on a vendor website?

It is a marketing phrase with no standard definition. A vendor using it is not necessarily doing anything wrong - they may have genuinely strong security. But the phrase itself confirms nothing. Certifications like SOC 2 Type II or ISO 27001 describe a verified standard; enterprise-grade does not.

What questions should I ask a vendor that the security page won’t answer?

Three useful ones: Has the vendor had a security incident in the past two years, and if so how did they notify customers? Who is the named contact for security concerns? Can they share a recent SOC 2 Type II report under NDA? How a vendor answers these - the speed and the specificity - tells you something the page cannot.

How do I know which AI tools are safe for sensitive business data?

No tool is risk-free, but meaningful signals include a current SOC 2 Type II or ISO 27001 certification, a clear data processing agreement with a model-training opt-out, and a vendor who answers direct security questions without deflecting. For regulated data like health records, you need a signed agreement from the vendor before you start.