Privacy Policy
The full detail on data use, providers, transfers, retention, and your rights.
Read the policyWhat we do today, in plain language, so your security team can start their review before the first call.
Compliance
This page describes our controls. It is not a certification. For our current compliance status, data processing agreements, and sub-processor details, email contact@stacklift.ai and tell us what your review requires.
The full detail on data use, providers, transfers, retention, and your rights.
Read the policyWhat our AI estimates can and can't tell you, including the limits of AI output.
Read the disclaimerThe security reviews and release checks we build into the software we deliver for you.
See delivery operationsAccounts and access
Staff access is set module by module, at no access, read, write, or admin level, so people reach only the areas their role needs.
Staff accounts can turn on authenticator-app codes with single-use backup codes. Authenticator secrets are encrypted, and backup codes are stored hashed.
Passwords are stored hashed. Repeated failed attempts lock the account for a set period, and sessions use HttpOnly, Secure cookies.
Sign-in events, AI requests, and changes to AI settings are logged, so they can be reviewed later.
Your data
The site enforces HTTPS with HSTS, and data in transit uses TLS 1.2 or later.
Data at rest is encrypted with AES-256. The platform is hosted on Vercel in the USA, with EU and UK transfers covered as set out in our Privacy Policy.
API keys and integration tokens we hold are encrypted with AES-256-GCM before they are saved.
Our lead forms use server-side rate limits, bot traps, and input validation.
Enquiry records are kept for up to 3 years from last contact. Client project data is kept for the engagement plus 7 years, for accounting and tax obligations.
Deletion requests are completed within 30 days, subject to legal exceptions. If a breach puts your rights at risk, we notify the relevant authority within 72 hours and affected people without undue delay.
AI and your data
AI speeds up scoping and delivery work. These are the limits we keep around it.
Your project inputs are processed by AI models at inference time. They are not used to train or fine-tune any AI model unless you give separate, explicit consent.
Depending on configuration, inputs may go to OpenAI, Anthropic, Google, Microsoft Azure OpenAI, AWS Bedrock, or self-hosted models.
Agents work on feature branches. They never push to the default branch or touch production secrets.
AI output is a draft and can be wrong. Every estimate is reviewed by a senior architect before it becomes a commitment, and engineers review agent output before it reaches production.
Common questions
No, not without your separate, explicit consent. Your project inputs are processed by AI models at inference time only.
Yes, ask us. Email contact@stacklift.ai with what your review requires, and we will tell you what is available for your project.
They remain your property. You grant us a limited license to process them for your estimate and any formal proposal that follows.
Ownership of delivered work is agreed in your Statement of Work and engagement agreement, so you can review it before you sign.
Agents work on feature branches and never push to the default branch or touch production secrets. Engineers own production changes.
Email contact@stacklift.ai with the details and put "Security" in the subject line.
The complete guide
Updated November 18, 2026 · Published by Danny Rathod, Founder, StackLift AI. Written with AI assistance; how we make content.
Short answer
A vendor security review checks how a supplier controls access, protects and deletes data, uses AI, and secures what it delivers. StackLift AI publishes its controls: role-based access, two-factor sign-in, encrypted credentials, audit logs, no AI training on your inputs without consent, and agents that stay on feature branches. This is a description of controls, not a certification.
A vendor security review is how a buyer decides whether a supplier can be trusted with its data and systems. It usually runs as a questionnaire, a document request and sometimes a call. Reviewers want evidence, not adjectives.
The questions fall into a few groups: who can access what, how data is protected in transit and at rest, how long it is kept and how it is deleted, how credentials and secrets are handled, how incidents are reported, and, increasingly, how AI is used. For software delivery, the review also covers the security of the code the vendor ships to you.
| Area | What reviewers ask |
|---|---|
| Access control | Who can reach which systems, and how is it limited and logged? |
| Authentication | How are accounts, sessions and second factors protected? |
| Data protection | Is data encrypted in transit and at rest, and where is it hosted? |
| Retention and deletion | How long is data kept, and how are deletion requests handled? |
| Secrets | How are API keys and tokens stored and rotated? |
| AI use | Which providers see our inputs, and is data used for training? |
| Secure delivery | How is the code reviewed, tested and released? |
| Incidents | How and when would we be told about a breach? |
Staff access is set module by module, at no access, read, write or admin level, so people reach only the areas their role needs. That is least privilege applied to a real system, and it is the first thing a reviewer wants to see.
Staff accounts can turn on two-factor sign-in with authenticator-app codes and single-use backup codes. Authenticator secrets are encrypted, and backup codes are stored hashed. Passwords are stored hashed, repeated failed attempts lock the account for a set period, and sessions use HttpOnly, Secure cookies. Sign-in events, AI requests and changes to AI settings are logged so they can be reviewed later.
The site enforces HTTPS with HSTS, and data in transit uses TLS 1.2 or later. Data at rest is encrypted with AES-256. The platform is hosted on Vercel in the USA, with EU and UK transfers covered as set out in the Privacy Policy. API keys and integration tokens we hold are encrypted with AES-256-GCM before they are saved.
Public lead forms use server-side rate limits, bot traps and input validation. Retention and deletion terms are published: enquiry records and client project data are kept for stated periods, deletion requests are completed within a stated window subject to legal exceptions, and a breach that puts your rights at risk triggers notice to the relevant authority and affected people. The Privacy Policy holds the exact terms.
| Control | Where to read it |
|---|---|
| Encryption in transit and at rest | The security page, data section |
| Stored credentials | The security page, data section |
| Hosting and transfers | The security page and the Privacy Policy |
| Retention and deletion | The security page and the Privacy Policy |
| Breach notice | The security page and the Privacy Policy |
Not without separate, explicit consent. Project inputs are processed by AI models at inference time. They are not used to train or fine-tune any model unless you agree to it separately. This is the first AI question most reviewers ask, so get it in writing from every vendor.
Depending on configuration, inputs may go to OpenAI, Anthropic, Google, Microsoft Azure OpenAI, AWS Bedrock or self-hosted models. Ask which of those apply to your project during scoping, because the choice affects where data goes. If data must stay inside your environment, the private LLM development page explains that option.
| Control | What it says |
|---|---|
| Not used for training | Inputs are processed at inference time and not used to train or fine-tune without separate consent |
| Named providers | Inputs may go to the listed providers depending on configuration |
| Limits on agents | Agents work on feature branches and never push to the default branch or touch production secrets |
| People make the call | Every estimate is reviewed by a senior architect, and engineers review agent output before production |
For the risks AI adds, see the glossary entries on prompt injection and AI guardrails.
A vendor can protect its own systems well and still ship insecure code. Review the delivery side separately. Our Delivery Operations page lists the practices that run on every sprint: security architecture review, authentication audit, an OWASP Top 10 check, secrets management with no hardcoded credentials, data encryption review, access control audit, and dependency and static scanning on pull requests.
The OWASP Top 10 is a widely used list of the most common web application risks, such as broken access control and injection. A review against it is a baseline, not a full penetration test. If your policy needs independent testing, say so during scoping so it is planned and priced.
Ask for documents, not assurances. A vendor with real controls can describe them concretely and tell you what it can and cannot share. Treat vague answers as an answer.
StackLift AI asks you to send your questionnaire with project details, and we tell you what we can share. This page describes our controls. It is not a certification, so ask for the current compliance status and the documents your review requires. The security review checklist and the vendor due diligence checklist give you a longer list.
Your project inputs remain your property. You grant StackLift AI a limited license to process them for your estimate and any formal proposal that follows. Ownership of delivered work is agreed in your Statement of Work and engagement agreement, so you can read it before you sign.
Check this point with every vendor. Ownership, licensing of third-party components and handover of documentation and access should all be explicit. See the software handover checklist for what a complete handover contains.
A public page is a summary. It cannot show how a control behaves under pressure, and it cannot replace your own review. Use it to shorten the questionnaire, then verify the controls that matter most to you.
If you find a security issue, email the team with the details and put Security in the subject line. For anything contractual, rely on the signed documents, not the website.
No, not without your separate, explicit consent. Your project inputs are processed by AI models at inference time only. They are not used to train or fine-tune any model.
No. Agents work on feature branches and never push to the default branch or touch production secrets. Engineers own production changes, and they review agent output before it ships.
Ask the team. Email what your review requires and we will tell you what is available for your project. The security page describes controls and is not a certification.
Depending on configuration, inputs may go to OpenAI, Anthropic, Google, Microsoft Azure OpenAI, AWS Bedrock or self-hosted models. Confirm which apply to your project during scoping.
Ownership of delivered work is agreed in your Statement of Work and engagement agreement. You can review those terms before you sign. Your project inputs remain your property.
API keys and integration tokens we hold are encrypted with AES-256-GCM before they are saved. Delivery practice also requires secrets management with no hardcoded credentials in the code we ship.
Email the team with the details and put Security in the subject line. Include what you saw and how to reproduce it.
Send it over with your project details, and we will tell you what we can share.