# Startup security checklist: the basics before enterprise customers ask

URL: https://pixelactstudio.com/resources/startup-security-checklist

> Startup security checklist: 43 yes or no checks on access, secrets, tenancy, backups, and logging, with what to fix before a security questionnaire.

A startup security checklist is a list of yes or no checks showing whether a young SaaS product has the security basics enterprise customers, investors, and questionnaires ask about. Pixelact's checklist has 43 checks in seven groups: accounts and access, secrets and dependencies, user authentication, authorization and tenant isolation, data protection and backups, logging and incident response, and questionnaire readiness.

It is written for founders, CTOs, and engineering leads at seed and Series A SaaS companies. Use it before your first enterprise deal sends a security questionnaire, before investor technical due diligence, or when a first version built fast, often with AI coding tools, starts holding real customer data. Our guide to [AI-generated code risks](https://pixelactstudio.com/blog/ai-generated-code-risks) covers that last situation in more depth.

Work through it with someone who can open the repository, the cloud console, and the auth provider settings. Answer each check for the product as it runs today, not as planned, and count a yes only when you could show the evidence, such as a setting, a test, a log, or a document. Partly done counts as no.

Then use the scoring notes at the end, because where the no answers fall matters more than how many there are.

> **What this checklist is not.** Passing every check does not make a product secure or compliant with SOC 2, ISO 27001, or GDPR. The checks cover common, well-documented risks and show you where to look first. Input validation and injection are outside this list, and the [AI-built app risk checklist](https://pixelactstudio.com/resources/ai-built-app-risk-checklist) covers them.

## Are company accounts, access, and laptops locked down?

- [ ] Is multi-factor authentication enforced, not just available, on company email, GitHub, your cloud console, domain registrar, and payment accounts? These accounts can reset every other password, push code to production, or move money. Enforce MFA at the organization level in Google Workspace or GitHub, and prefer passkeys or hardware keys over SMS codes, which can be phished or lost to a SIM swap.
- [ ] Does every leaver lose access to email, GitHub, cloud, hosting, databases, and shared vaults on their last day, following a written offboarding list? A contractor who left months ago can still have GitHub or cloud access if nobody owned removing it. Enterprise questionnaires ask for this procedure by name, and a written list is the evidence.
- [ ] Does each production cloud account have at least two named owners, with the root or owner login protected and kept for emergencies? An AWS root user or a sole Google Cloud owner tied to one founder's email is a single point of failure and a high-value target. Two owners mean a lost phone or a departure doesn't lock the company out of its own infrastructure.
- [ ] Do engineers work with read-only production access by default, and get write access only for a specific task? Standing admin access means one stolen session can drop a table or export every customer record. Least privilege also limits honest mistakes, such as running a migration against the wrong database.
- [ ] Are shared credentials kept in a team password manager rather than in chat, a spreadsheet, or a notes app? A password pasted into Slack stays searchable for everyone who joins later and can't be traced or rotated cleanly. A manager such as 1Password or Bitwarden shows who has access and makes rotation a single change.
- [ ] Do company laptops use full-disk encryption, an automatic screen lock, and automatic OS updates? A lost laptop can carry a signed-in GitHub session, cloud credentials, and a local database copy. FileVault or BitLocker with a short lock timer is a standard questionnaire answer and quick to roll out across a small team.

## Are secrets kept out of code, and dependencies patched and pinned?

- [ ] Have you scanned the full git history, not only the current code, for API keys, database URLs, and private keys? A key deleted in a later commit is still readable in every clone and fork. [GitGuardian's State of Secrets Sprawl 2026 report](https://www.gitguardian.com/state-of-secrets-sprawl-report-2026) tracked valid secrets detected in public repositories in 2022 and found that 64% were still active four years later.
- [ ] Has every secret that reached git, a ticket, or a chat been rotated at the provider, not only deleted from the code? Deleting a key from the code does not revoke it, so anyone holding a copy can keep using it. Our teardown of [exposed API keys](https://pixelactstudio.com/blog/exposed-api-keys) shows the rotation step by step.
- [ ] Do production secrets live in your host's secret store or a secrets manager, with different values from staging and development? When environments share keys, a leaked staging `.env` file opens production. Separate values in AWS Secrets Manager, Doppler, or your platform's encrypted environment variables keep a laptop leak contained. Our open source tool [EnvSift](https://pixelactstudio.com/work/envsift) turns a `.env` file into a `.env.example` with the values removed, inside the browser, so the repository lists variable names without holding secrets.
- [ ] Do your web and mobile app bundles contain only publishable keys, with nothing secret read from a `NEXT_PUBLIC_`, `VITE_`, or `EXPO_PUBLIC_` variable? Build tools write public-prefixed variables into the shipped JavaScript as plain text that anyone can download. An LLM key, a Stripe secret key, or a database password there can be copied from the browser's developer tools or unpacked from the app.
- [ ] Does CI or a push protection rule block commits and pull requests that add a secret? A one-time cleanup decays as soon as someone pastes a new key. Gitleaks in CI or GitHub push protection stops the next one before it reaches the main branch.
- [ ] Do dependency alerts from Dependabot, Renovate, or a similar tool reach an owner who acts on critical ones within days? Published vulnerabilities come with public details, so an unpatched package tells an attacker where to look. Software supply chain failures rank third in the [OWASP Top 10:2025](https://top10.owasp.org/2025).
- [ ] Is the lockfile committed, and does CI install from it exactly with `npm ci` or `pnpm install --frozen-lockfile`? Without a frozen install, each build can resolve different versions, including a newly published malicious release. The lockfile makes the code you tested the code you ship, and it shows reviewers every package change in a pull request.

## Is user sign-in safe from account takeover?

- [ ] Is sign-in built on a maintained auth provider or library rather than hand-written password and session code? Password storage, reset tokens, session rotation, and rate limits each have well-known failure modes. Auth0, Clerk, Supabase Auth, and Better Auth already handle them, while custom code has to get every one right.
- [ ] Are passwords hashed with Argon2id, scrypt, or bcrypt, and never stored in plain text, logged, or hashed with a fast function such as SHA-256? If the users table leaks, fast hashes can be cracked in bulk on commodity GPUs. Slow, salted hashes make each guess expensive, which buys time to force password resets.
- [ ] Are sign-in, sign-up, and password reset rate-limited per account and per IP address? Without limits, credential stuffing scripts try email and password pairs leaked from other sites against your login at high speed. Limits on reset and sign-up also stop email floods that damage your sending reputation.
- [ ] Do password reset and magic links expire quickly, work only once, and stop working after the password changes? A reset link that never expires turns an old email in a shared or forwarded inbox into an account takeover. Single-use tokens also stop a link that leaked through a proxy log from being replayed.
- [ ] Are session cookies set with `HttpOnly`, `Secure`, and `SameSite=Lax` or `Strict`, and are sessions revoked on sign-out and password change? Without `HttpOnly`, one cross-site scripting bug can read the session token. Without revocation, a stolen session keeps working after the user changes their password to lock the attacker out.
- [ ] Can customers turn on MFA, and can an organization admin require it for every member? An admin account in your product can see and export a customer's whole workspace, so enterprise questionnaires ask about MFA for end users directly. Most auth providers support authenticator apps and passkeys through configuration rather than new code.

## Can one customer see another customer's data?

- [ ] Is authorization checked on the server for every API route, server action, and background job, not only by hiding buttons in the UI? Anyone can call your API directly from the browser's network tab or a script, so a hidden button protects nothing. Broken access control is first in the OWASP Top 10:2025.
- [ ] Does every query on tenant data scope by the organization from the authenticated session, never by an ID sent in the URL or request body? If the server trusts an `org_id` from the client, a user can change one number and read another company's records. This insecure direct object reference often passes manual testing, because each tester only uses their own account.
- [ ] If the browser queries the database directly, as with Supabase or Firebase, is row-level security or a security rule enabled on every table and tested? With Postgres row-level security off on a table the API can reach, the public key shipped in your frontend can read every row through the auto-generated API. A policy that scopes rows to the user's organization closes that path.
- [ ] Does an automated test prove that a user in one organization cannot read or change another organization's data? Tenant isolation breaks quietly when someone adds an endpoint or rewrites a query. A test that signs in as one tenant and requests another tenant's records catches the regression in CI, before a customer does.
- [ ] Are roles and permissions defined in one place, with invites, billing changes, and data exports limited to admin roles? When permission checks are copied into each handler, one forgotten copy gives every member admin powers. A single policy function or middleware is easier to review, test, and describe in a questionnaire.
- [ ] Do internal admin and support tools require staff MFA and record every customer record a staff member views or changes? Admin panels can read every tenant by design, which makes them the most valuable target in the product. Enterprise buyers ask who at your company can see their data and how you would know.

## Is customer data encrypted, backed up, and restorable?

- [ ] Is every connection encrypted with TLS 1.2 or later, including connections to the database, with HSTS set on your domains? Questionnaires ask for encryption in transit on every hop, not only the public site. HSTS stops browsers from falling back to plain HTTP, where traffic can be read or changed on a shared network.
- [ ] Is encryption at rest turned on for the database, backups, and file storage, and do you know who manages the keys? Most managed databases and object stores offer encryption at rest, often on by default, but questionnaires ask you to confirm it per service and name the key owner. Answering “we think so” slows the review.
- [ ] Does the production database have automated backups or point-in-time recovery, with copies kept in a separate account or project from production? A backup in the same account can be deleted by the same stolen credential or the same mistaken script. Enterprise buyers ask how you would recover from ransomware, an accidental deletion, or a lost cloud account.
- [ ] Have you restored a backup into a fresh database in the last three months and timed it? A backup nobody has restored is an assumption. The timed restore gives you real figures for recovery time and data loss, which questionnaires call RTO and RPO.
- [ ] Do you know every place customer personal data is stored, including logs, analytics, error tracking, and staging copies? Deletion requests under GDPR and data clauses in customer contracts cover every copy, not only the main database. A one-page data map is also the first thing a reviewer asks for when a questionnaire turns to privacy.
- [ ] Does staging run on synthetic or masked data rather than a raw copy of production? Staging usually has weaker access controls and more people with access. A full production copy there turns a minor staging leak into a customer data breach.
- [ ] Are uploaded files stored in private buckets and served through signed URLs that expire? A public S3 or storage bucket with predictable paths exposes every invoice or contract anyone uploads. Signed URLs that expire after minutes limit a leaked link to one file for a short time.

## Would you notice a security incident, and who would respond?

- [ ] Do you keep an audit log of sign-ins, permission changes, data exports, and admin actions, retained for a set period where the app itself cannot delete it? Without one you can't answer the first question of any incident: what did the attacker touch? Buyers also ask for the retention period, and logs that live only on the server they describe disappear with it.
- [ ] Are passwords, tokens, API keys, and sensitive request bodies kept out of logs and error reports? Logging an `Authorization` header copies a live credential into a third-party tool that more people can read. Sentry and most log platforms support scrubbing rules, but someone has to configure them.
- [ ] Does an alert reach a named person when error rates spike, sign-ins fail in bulk, or a new admin account appears? Logs nobody reads only help after the damage is done. Security logging and alerting failures have their own entry in the [OWASP Top 10:2025, at A09](https://top10.owasp.org/2025/A09_2025-Security_Logging_and_Alerting_Failures).
- [ ] Is there a one-page incident response plan that names who leads, who decides on customer notice, and how to reach your host and auth provider? Under [GDPR Article 33](https://gdpr-info.eu/art-33-gdpr/), a processor must tell the controller about a personal data breach without undue delay, and the controller must notify the supervisory authority within 72 hours where feasible. Your business customers are usually the controllers, so they will need facts from you fast. The first hour of an incident is the wrong time to look up support contacts and contract deadlines.
- [ ] Can a security researcher find where to report a vulnerability, such as a security contact address or a `security.txt` file? Without a published channel, reports land in a support queue or on social media. A `/.well-known/security.txt` file, defined in [RFC 9116](https://www.rfc-editor.org/rfc/rfc9116), needs only a `Contact` and an `Expires` field to tell finders where to write.

## Are you ready for a security questionnaire?

- [ ] Do you keep a current list of the subprocessors that handle customer data, with what each one receives? Questionnaires and data processing agreements ask for subprocessors by name: hosting, database, email, analytics, error tracking, and any LLM provider. Building the list during a deal means guessing under a deadline.
- [ ] Can you say which region each system stores and processes customer data in? European buyers often ask about data residency and transfers outside the EU. An answer such as “the database is in Frankfurt, but error tracking and the LLM provider process data in the US” is only possible if someone has checked.
- [ ] Do short written policies for access control, incident response, backups, and vendors match what the team actually does? Questionnaires ask whether each policy exists and when it was last reviewed. A two-page policy that matches practice holds up when a reviewer asks for evidence, while a downloaded template describing controls you lack does not.
- [ ] Do you keep a reusable answer library with one owner, built on a standard question set such as the CSA CAIQ? Enterprise questionnaires repeat the same questions in different wording. The [CAIQ v4.1](https://cloudsecurityalliance.org/artifacts/star-level-1-security-questionnaire-caiq-v4-1) from the Cloud Security Alliance is a public set of yes or no questions mapped to its Cloud Controls Matrix, so answering it once gives you a base for later forms.
- [ ] Do your answers describe what is true today, with each gap stated next to a planned fix date? A yes that isn't true yet can be written into the contract as a commitment and surface at renewal or after an incident. A stated gap with a dated plan is something a buyer can weigh, while a vague yes is not.
- [ ] Do you know when your largest prospects will expect SAML single sign-on, a SOC 2 report, an ISO 27001 certificate, or a recent penetration test? None of these can be produced in the last week of a deal, and a SOC 2 Type II report covers an observation period of several months. Knowing early lets you plan the work and the cost, and check whether your auth provider includes SAML on your plan.

## How do you score the startup security checklist?

Score the startup security checklist by where the no answers fall, not by the total. One no in tenant isolation can expose every customer at once, while a missing security contact only slows a deal down.

### Which security gaps are urgent?

Treat these as urgent and fix them within days, before more customer data arrives.

- A live secret in git history, a client bundle, or a chat that has not been rotated.
- Any route, query, or table where one customer can read another customer's data, including row-level security turned off.
- Passwords stored in plain text or with a fast hash, or customer files in a public bucket.
- Company email, GitHub, or a cloud owner account without enforced MFA, or a former team member who still has access.
- Backups that have never been restored, or that sit in the same account as production.

The other authentication, access, and data protection checks come next, within a few weeks. Gaps in logging, policies, and questionnaire answers are real but less urgent. Plan them over the next one to two months, and close them before a large prospect starts its security review.

### When does an outside security review make sense?

An outside review makes sense when several urgent checks came back no, when nobody on the team could answer a check, or when an enterprise review or investor diligence is a few weeks away.

Pixelact's [software audit](https://pixelactstudio.com/services/software-audits) is about 3 to 5 days of review across your code, architecture, infrastructure, and workflow, including the checks on this list that live in code and configuration. It ends in a written report that ranks each finding by harm and effort, a build path with rough costs, and a walkthrough call. It is a fixed fee from $2,500 and is not a penetration test or a certification.

Our guide to [what a software audit covers](https://pixelactstudio.com/blog/what-a-software-audit-covers) explains the report in detail, and you can ask for the sample report we made from one of our own products when you [request an audit](https://pixelactstudio.com/request-an-audit).

## Frequently asked questions

### What security basics do enterprise customers expect from a startup?

Enterprise buyers usually expect enforced MFA, secrets kept out of code, server-side authorization, tested tenant isolation, restore-tested backups, audit logs, and an incident plan. They check through a security questionnaire, either their own or a standard set such as the Shared Assessments SIG or the CSA CAIQ, plus a data processing agreement. Larger buyers may also ask for evidence, such as a screenshot of MFA enforcement, and for a SOC 2 report or a penetration test summary.

### Does passing this checklist make us SOC 2 or GDPR compliant?

No. This checklist covers common, well-documented security risks in a SaaS product and the questions buyers ask about them. A SOC 2 report comes from an independent CPA firm's examination, and ISO 27001 certification comes from an accredited certification body. GDPR duties depend on the personal data you process and whether you act as controller or processor, so they need legal advice. A strong result here makes that later work easier, but it doesn't replace it.

### How do you prepare for a security questionnaire without a security team?

Give one person ownership and one shared folder. Before the first form arrives, collect dated evidence: screenshots of MFA enforcement, your subprocessor list, data regions, encryption settings, and your last restore test result. Write short policies that match what you do, then answer a standard set such as the CSA CAIQ once and reuse those answers. State each gap honestly, with a planned fix date. Expect the first questionnaire to take the longest.

### When does a startup need a penetration test?

Book a penetration test when a customer contract, an enterprise buyer, or an investor asks for one, or before launching a feature that handles payments or sensitive data. Fix the basics on this checklist first, so the testers' paid days go to harder problems. Ask for authenticated testing with accounts in two organizations, so testers can try cross-tenant access. Pixelact doesn't run penetration tests, but our software audit lists what to fix before you book one.

### We built our first version with AI coding tools. Where should we start?

Building the first version with AI coding tools was a sensible way to reach users quickly. Start with the checks on secrets and on tenant isolation, because a coding assistant can't know your tenancy rules or which keys must stay on the server unless someone tells it. Then check backups and logging. The [AI-built app risk checklist](https://pixelactstudio.com/resources/ai-built-app-risk-checklist) covers input validation, migrations, error handling, tests, and AI API spend.