Skip to content

// Frequently asked questions

Frequently asked questions about pentest and SaaS security

Every answer published across RET guides, comparisons and tool pages, in one place. Each answer links to the page that covers the topic in full.

Guides

Security checklist for AI-built SaaS

Does this checklist replace a pentest?

No. It organizes the first review and helps you prioritize. When a signal touches access, paid plans, data, or a B2B sale, the next step is a human review (Risk Review) or a scoped manual pentest.

Does it work for apps built in Lovable, Bolt, v0, or Replit?

Yes. The checklist inspects real product flows (login, billing, data, uploads), not the tool that generated the code. It applies to any AI-built SaaS.

My SaaS is still small. Does it make sense to review now?

Yes, if you already have users, billing, customer data, or a commercial promise. The earlier you separate signal from noise, the cheaper the fix.

Read the full page

MCP tool poisoning: how to prevent it

Are prompt injection and tool poisoning the same thing?

They are cousins. Prompt injection is the technique of hiding instructions in text; tool poisoning applies it via an MCP tool description or return to hijack the agent. The defense overlaps: boundaries, least privilege, and distrust of external context.

Can I fix it just by using a better model?

No. This is a permission-architecture problem, not a model-quality one. Even a strong model runs a bad action if it has broad permission and untrusted context.

Does OWASP cover this?

Yes. The OWASP MCP Top 10 and the Agentic Security Initiative name tool poisoning, context spoofing, and broad permissions. They serve as a map to prioritize the review.

Read the full page

Prompt injection in code agents

Doesn’t the system prompt protect against this?

It helps, but it is not armor. Instructions hidden in context can compete with the system prompt. The real protection is least privilege + diff review + secrets kept out of context.

Does this apply to Cursor, Claude Code, Codex, and Copilot?

It applies to any agent that reads context and writes code. The tool name changes, not the principle: untrusted content + broad permission = risk.

How do I test whether my flow is exposed?

Simulate a malicious snippet in a place the agent reads (issue, comment, dependency) and see if its behavior changes. If it does, reduce permission and add human review.

Read the full page

IDOR in Stripe checkout

Does using UUIDs instead of sequential ids fix IDOR?

It reduces guessability, but it is not a fix. IDOR is a missing ownership check. If the route does not check who may see the object, a leaked id (in a log, referer, screenshot) still exposes the data.

Can the Stripe success page grant access?

Not as a source of truth. The session_id in the success URL does not prove payment. Confirm via the signed webhook or by querying Stripe server-side before unlocking a plan.

How do I know if my checkout has IDOR?

Create two accounts, place an order in each, and try to access one’s object using the other’s session (swapping ids). If data comes back, it is IDOR. On a revenue flow, validate it with a manual test.

Read the full page

Stripe webhook security

Can I unlock access on the success page?

Not as a source of truth. For async methods (Pix, boleto), the payment may not have settled. Unlock via the confirmed webhook; use the success page only to guide the user.

Why does signature verification fail sometimes?

Almost always because the body was parsed (or modified) before the check. The signature needs the raw body exactly as received.

Do I really need idempotency?

Yes. Providers resend events by design. Without idempotency, you risk double delivery, wrong credit, or access unlocked more than once.

Read the full page

Multi-tenant isolation in SaaS

Is database RLS enough on its own?

It is a strong safety net, but the ideal is layered defense: application scope + database RLS. RLS catches what the application forgets, and the application avoids depending on the database alone.

How do I test isolation without breaking production?

Create two test tenants and try, with one’s session, to read and write the other’s resources. An automated test expecting denial on cross-tenant access becomes a permanent regression guard.

Does it apply to small SaaS too?

It applies from the second customer on. Introducing isolation early is cheap; leaking data between customers is expensive and hard to reverse.

Read the full page

How to review login and authentication in AI-built SaaS

Is MFA mandatory for small SaaS?

For admin accounts, treat it as essential: it is the most valuable target. For regular users, offer it as an option. A compromised admin is too expensive to depend on a password alone.

Does the login error message really matter?

It does. "E-mail not found" versus "wrong password" hands an attacker a list of valid e-mails for brute force and phishing. Use a generic message.

Does a correct login guarantee security?

No. Many incidents happen after login, in authorization: an authenticated person reaches another customer’s data or an admin action. Review authorization separately from login.

Read the full page

Security by AI build tool: which to review first

Is an app built in Lovable/Bolt/v0/Replit insecure by nature?

No. The generated code can be good. The risk is shipping sensitive flows without reviewing the boundary (authorization, isolation, secrets) the generator does not close automatically.

Do I need to migrate tools to be secure?

Usually not. The review inspects the product flow, not the code’s origin. Migrating without reviewing just moves where the same mistake lives.

I use Supabase; does RLS solve everything?

RLS helps a lot, but only if it is enabled and correct. Many projects leave policies permissive or disabled. Confirm the rules and combine them with application-level checks.

Read the full page

Secure file upload in SaaS

Is checking the file extension enough?

No. Extension and content-type are declared by the client and forgeable. Validate the real type by magic bytes on the server and define an allowlist of what is accepted.

What is SSRF in upload?

It happens in "import by URL": the server fetches the address the user sent. Without an allowlist, it can fetch an internal service or the cloud metadata endpoint and leak credentials.

Can I serve the file straight from the public folder?

Avoid it. Store in a private bucket and serve via a temporary signed URL, with tenant scope. A public folder invites file IDOR and improper execution.

Read the full page

Credential stuffing and MFA on the admin panel

Does a strong password solve credential stuffing?

It helps against guessing, but not against reuse: if the same password leaked elsewhere, it is already known. The core defense is MFA, especially on admin.

Is SMS MFA enough?

It is better than nothing, but SMS is vulnerable to SIM swap and interception. Prefer an authenticator app (TOTP) or a security key, especially for admin.

How do I know if I am under credential stuffing?

Spikes in login failures, many attempts across different e-mails from the same IP/range, and successes from unusual locations. Without rate limiting and logs, the attack goes unnoticed, start by measuring.

Read the full page

Security incidents in AI-built SaaS in 2026

My app is small, am I really a target?

The 2026 cases show size does not protect: Moltbook leaked 1.5 million tokens within 72h of life, and the RedAccess study found leaks in thousands of small apps. Automated scanners sweep new apps at scale; the target is the exposed data, not fame.

Is this the AI tool’s problem or mine?

Both. The tool accelerates and leaves access control loose by default, but the app and the data are yours, so the correct policy is the publisher’s responsibility. The good news is most of these flaws are quick to fix once you know where to look.

How do I know if I am exposed?

By testing the access flows: what the public key reaches with no login, whether a swapped id returns another customer’s data, whether an endpoint responds with no auth. The Promptbook guides that check and Risk Review does the human read when the signal is serious.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence. RET has over 100 real vulnerabilities reported with responsible disclosure.

Read the full page

How to pass a security due diligence

Do I need a pentest before the round?

Not mandatory, but it helps a lot. Arriving at due diligence with a pentest report and the fixes done changes the dynamic: instead of the buyer finding problems, you show you already solved them. That protects valuation and timeline.

What most kills a deal on the security side?

Cross-tenant leakage (multi-tenant), an exposed secret in the front-end, and the total absence of any testing. The first two are concrete findings; the third is a maturity signal that scares on its own.

Can I start cheap?

Yes. Risk Review reads 1 to 3 sensitive flows and tells you what to fix before due diligence. If the deal requires a formal report, the Manual Pentest delivers the evidence in the expected format.

Who runs it at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence.

Read the full page

How to answer a customer security questionnaire

Can I answer that I have a control I have not implemented yet?

No. Inventing a control is the worst path: it shows up in the first audit and burns trust. Better to admit the gap with a plan and a fix timeline; a mature buyer respects that more than an all-green, false spreadsheet.

What unblocks a questionnaire most?

Independent testing (a third-party pentest report) and proof of tenant isolation. Those are the two things the buyer most wants to see and that your word alone does not back.

Do I need a pentest just to answer the questionnaire?

It helps a lot and is sometimes required. The report answers a whole section at once and counts as third-party proof. If budget is tight, Risk Review covers the most-asked flows at a lower cost.

Who runs it at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence the customer accepts.

Read the full page

SSRF in SaaS: how to find and prevent it

Is SSRF really serious? The server just fetches a URL.

It is serious because the server has access the attacker does not: the cloud metadata endpoint returns the instance credentials, and the internal network usually has no auth. An SSRF becomes credential theft or access to an internal service.

Is validating the URL string enough?

No. The attacker bypasses it with DNS that resolves to an internal IP, with redirects, or with alternate IP notation. The correct defense resolves the host and validates the resolved IP before connecting, and revalidates on each redirect.

How do I test without risk?

Point the feature (preview, import, webhook) at the metadata IP and at a test internal service. If data comes back, it is SSRF. The Promptbook guides that check across URL-fetching features.

Read the full page

Supply chain security: npm dependencies in 2026

What was Shai-Hulud?

A self-propagating npm worm, detected in 2025 with a new wave in 2026, that compromised a maintainer, stole cloud and npm tokens, and republished itself into more packages using those tokens. It ran on preinstall and spread through CI/CD, where tokens are plentiful.

How do I protect myself without reading every package?

By shrinking the blast radius: a bump cooldown (holds the new version for days), a pinned lockfile, a CI token with minimal scope and rotation, and disabling lifecycle scripts when possible. It does not eliminate risk, but cuts most cases.

Does SCA (Snyk/Dependabot) solve it?

It helps a lot on the dependency tree (known CVE, rotten version), and you should keep it on. But it does not see the blast radius in your CI or whether a compromised token reaches production; that is application and infra review.

Who runs it at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope, focusing on blast radius and secret isolation.

Read the full page

Insecure deserialization RCE: the React2Shell case

Does React2Shell affect me if I use Next.js?

If you ran a vulnerable version of React Server Components / Next.js in production, yes, including a default create-next-app app. The fix came as a patch; updating was the urgent step. Apps on fixed versions are no longer affected by that specific CVE.

Is insecure deserialization common in AI-built SaaS?

The most severe case (React2Shell) was in the framework, not the dev’s code. But in your code the risk exists whenever you reconstruct an object from untrusted input. Using schema-validated JSON instead of formats that recreate arbitrary types closes most of it.

Is updating the framework enough?

It closes the specific CVE and is the first step. But the lesson holds going forward: do not deserialize untrusted data in your code, run with least privilege, and monitor the version against advisories. RET reviews that set and the blast radius.

Read the full page

Broken access control: the OWASP #1 risk in SaaS

Is broken access control the same as IDOR?

IDOR is a FORM of broken access control (swapping an id and reaching someone else’s resource). Broken access control is the larger family, which includes privilege escalation, forced browsing and front-end-only checks. IDOR is the most common.

Why is it OWASP #1?

Because it is easy to introduce (forgetting an ownership check), hard to see in the demo (it happens after login) and high impact (it leaks customer data, breaks the multi-tenant promise). In AI-built SaaS, it is the dominant flaw.

Does strong authentication solve it?

No. Authentication (login, MFA) proves who you are; broken access control is an authorization failure (what you can do after logging in). A perfect login with loose authorization still leaks data.

How do I test it quickly?

Create two accounts, do an action in each, and try, with A’s session, to read/edit B’s resource (swapping ids). If data comes back, it is broken access control. Risk Review does that sweep in depth.

Read the full page

CSRF in SaaS: what it is and how to prevent it

Does SameSite alone solve CSRF?

SameSite=Lax/Strict cuts most classic CSRF and is the cheapest defense to turn on. But for very sensitive actions, combine it with an anti-CSRF token and a server-side Origin check: layered defense is the secure standard.

Are CSRF and XSS the same?

No. XSS injects a script into your page (runs code in the victim’s browser). CSRF injects nothing; it makes the victim’s browser fire a legitimate request using their session. Different defenses.

Does an API with a header token need CSRF protection?

If authentication is by a header token (not an automatic cookie), classic CSRF does not apply, because the browser does not attach the token by itself. CSRF lives where the session rides on an automatic cookie.

How do I test it?

Build a test page that fires your app’s sensitive action and open it logged in on another tab. If the action happens with no token/origin check, it is CSRF. The Promptbook guides that check.

Read the full page

XSS in AI-built SaaS

Doesn’t React protect me from XSS automatically?

React escapes by default when you render text in JSX, which covers most cases. XSS enters when you bypass that, mainly with dangerouslySetInnerHTML on user-supplied content. That is where the protection is lost.

Which XSS is the worst?

Stored XSS: the malicious script is saved (in a comment, a profile) and runs in the browser of everyone who sees that content. A single poorly escaped field can compromise many users.

Does CSP replace escaping input?

No. Escaping input is the primary defense; CSP is the second layer, limiting the damage if a script gets through. Use both: escaping prevents, CSP contains.

How do I test it?

Insert a simple test payload into a field that is displayed later and see whether it executes instead of appearing as text. If it executes, it is XSS. The Promptbook guides that check on the spots that render HTML.

Read the full page

CSP (Content Security Policy) in SaaS

Does CSP replace escaping input against XSS?

No. Escaping input is the primary defense that prevents XSS; CSP is the second layer that contains the damage if a script gets through. The two together are the secure standard; either one alone leaves a gap.

Isn’t unsafe-inline necessary?

Almost never, and it greatly weakens the CSP (it allows any inline script, including one injected by XSS). The modern alternative is a per-request nonce with strict-dynamic, which executes only the legitimate script.

Does report-only CSP protect?

It blocks nothing; it only reports violations. It is great for discovering what an enforcing CSP would break before turning it on, but leaving it in report-only forever contains no attack. The goal is to reach enforce.

How do I know if mine is strong?

Check whether it uses a nonce (not unsafe-inline), whether default-src/object-src/base-uri are restrictive, and whether connect-src limits destinations. The Promptbook guides that quick audit.

Read the full page

SaaS security for fintech

Do I need PCI-DSS for my fintech?

It depends on how you touch card data. If you tokenize via a certified provider and do not store the PAN, the scope drops a lot. RET does not issue PCI (that is a QSA), but tests the technical security of the flows PCI and the partner ask about.

What is the worst fintech incident?

Improper access to another customer’s transaction/balance (IDOR) and a forged or double-processed payment webhook. Both touch money directly and become a regulatory incident, not just a bug.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence a financial partner accepts.

Does it work for AI-built fintech?

It does. The review looks at the money and data flows (transaction, webhook, KYC, isolation), not the tool that wrote the code.

Read the full page

SaaS security for healthtech

Is health data different from other data under privacy law?

Yes. It is SENSITIVE personal data, with a stricter legal basis and reinforced protection. A health-data leak carries much greater legal and reputational weight; isolation and access control must be airtight.

What is the worst healthtech incident?

A patient (or a clinic) being able to see another’s record/exam (an isolation failure or IDOR). It is exposure of intimate sensitive data, hard to recover in trust.

Does RET handle medical-council compliance?

The formal part (medical council, documentary privacy compliance) is your legal team’s/DPO’s. RET tests the technical security that underpins it: patient isolation, role-based access, clinical-data protection.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence.

Read the full page

SaaS security for e-commerce and marketplace

What is the most common fraud in AI-built e-commerce?

Manipulating price/discount/coupon on the client (AI validates on the front, not the server) and IDOR in orders. Both are cheap to exploit and expensive to ignore; price and coupon must be decided on the server.

What changes in a marketplace?

Seller isolation adds on: one store must not see another’s orders, revenue or customers. It is the same cross-customer leak risk, with contract weight between the stores.

Do I need PCI?

It depends on how you touch cards. Tokenizing via a certified provider greatly reduces the scope. RET does not issue PCI, but tests the technical security of checkout and payment that PCI asks about.

Read the full page

SaaS security for edtech

Is minors’ data treated differently under privacy law?

Yes. It gets reinforced protection for children and teenagers, with the minor’s best interest and, in many cases, a parent’s consent. That weighs on the access and retention design of your edtech.

What is the worst edtech incident?

A teacher or student reaching another class/school’s data (a multi-level isolation failure or IDOR), or leaking a minor’s data. It is sensitive because it involves children, with high legal and reputational weight.

Does RET handle parental consent?

The formal part (policy, consent) is your legal team’s/DPO’s. RET tests the technical security that underpins it: school/class/student isolation, role-based access and minors’ data protection.

Read the full page

SOC 2 for SaaS

Does RET give me the SOC 2 certificate?

No. SOC 2 is an attestation issued by an audit firm (CPA). RET delivers the technical security proof (pentest, Risk Review) that underpins the report’s Security criterion.

Does SOC 2 require a pentest?

The framework does not say "pentest" in those words, but the Security criterion asks for vulnerability management, and penetration testing is the evidence most accepted by auditors and customers. In practice, almost every SaaS SOC 2 includes a pentest.

Type I or Type II?

Type I attests control design at a point in time; Type II attests operation over a period. Enterprise customers usually require Type II. RET’s technical evidence serves both.

Who runs the pentest at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence the auditor accepts.

Read the full page

ISO 27001 for SaaS

Does RET certify me for ISO 27001?

No. Certification is issued by an accredited certification body. RET delivers the technical evidence for the Annex A controls that depend on testing (pentest, Risk Review).

Does ISO 27001 require a pentest?

Control A.8.8 (technical vulnerability management) and the secure-development controls are, in practice, met with security testing. A pentest is the most direct evidence of those controls for the auditor.

How is ISO 27001 different from SOC 2?

ISO 27001 certifies a management system (a process) via an accredited body; SOC 2 is an attestation of controls by an audit firm. RET’s technical evidence serves both.

Who runs the pentest at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence.

Read the full page

LGPD for AI-built SaaS

Does RET make my SaaS LGPD-compliant?

RET covers the technical part (Art. 46): isolation, access control, data protection, proven with evidence. The legal part (legal basis, policy, DPO, ANPD) is your legal team’s. LGPD compliance is the sum of both.

Is a data leak a technical or legal problem?

Both. The cause is usually technical (IDOR, isolation failure, exposed secret), and the consequence is legal (ANPD notification, fine, harm). RET attacks the technical cause before it becomes the consequence.

Does sensitive data change anything?

It does. Health, biometric and minors’ data get reinforced protection (Art. 11 and 14). Isolation and access control must be airtight, and technical review weighs more on those flows.

Read the full page

PCI-DSS for SaaS that touches payments

Does RET make me PCI compliant?

RET covers the technical part (Requirement 11 testing, security of payment flows). Formal validation (AoC, ROC, SAQ) is a QSA’s or self-assessment’s job. RET is not a QSA.

Do I need PCI if I use Stripe?

You still have PCI obligations, but tokenizing via a certified provider and not storing the PAN greatly reduces scope (usually to a simpler SAQ). The flows that touch payments still need to be secure.

Does PCI require a pentest?

Yes. Requirement 11 asks for regular security testing, including internal and external pentests of the card data environment. It is mandatory evidence, not optional.

Read the full page

The annual pentest required by Brazil’s Central Bank

Does a test by the internal team count?

As a complement. Article 22-A asks for a test carried out with independence and impartiality by an individual or specialized company hired for that purpose.

How often?

At least yearly. A major system change or a new exposed surface usually justifies an off-cycle test.

Does a small fintech need it?

If it is authorized by the Central Bank or is a payment institution, yes. Size is not an exception in Article 22-A.

Who runs the test at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence.

How much does it cost?

The 25-hour Sprint costs R$9,500, delivered within 7 business days, with an NDA the same day and a second check of your fixes with no limit for 90 days.

What if the test finds no flaw?

We refund 50% of the price. The other 50% pays for the hours executed, the report of what was tested and the attestation letter, which you can show to customers and auditors. It applies when the final report brings no flaw with a severity score in the agreed scope.

Read the full page

Pentest for fintechs

How much does a fintech pentest cost?

The Sprint costs R$9,500: 25 hours of manual testing on one web application, API or mobile app (Android or iOS), delivered within 7 business days. Larger scopes range from R$15,000 to R$38,000.

Does it meet Article 22-A?

It covers the technical part Article 22-A asks for: an independent specialist hired for the purpose, documented vulnerabilities and an action plan. Compliance and the annual report remain the institution’s responsibility; RET is not an auditor.

Do you test in production?

Only what the authorization form lists, inside the agreed window, with no denial of service and no change to real data.

How do I pay?

50% by PIX at signature and 50% on report delivery, or in up to 3 interest-free installments.

Read the full page

PIX pentest

How much does it cost?

The 25-hour Sprint costs R$9,500 for one web application, API or mobile app (Android or iOS).

Do you run real transactions?

Only with test accounts and amounts authorized in the form, never with customer money.

Does it meet Article 22-A?

For an institution under CMN 5.274 or BCB 538, the test delivers the technical part: independent execution and a report with flaws and an action plan. Compliance remains the institution’s responsibility.

Read the full page

API pentest

How much does an API pentest cost?

The Sprint costs R$9,500: 25 hours of manual testing on one API, delivered within 7 business days. A large API or several APIs range from R$15,000 to R$38,000.

Do you test GraphQL?

Yes, inside the authorized scope, including queries that expose another customer’s fields and nested queries with no limit.

Do I need to send documentation?

It helps. With a Postman collection or OpenAPI the test covers more in less time, in grey-box mode.

Read the full page

E-commerce checkout pentest

How much does it cost?

The 25-hour Sprint costs R$9,500 for one store’s checkout and order flows.

Do you place real orders?

Only with test cards and accounts authorized in the form, never touching customer orders.

My store runs on a hosted platform. Is it worth it?

Yes, for what is yours: scripts, apps and integrations installed in checkout. The platform itself is out of scope unless it authorizes the test.

Read the full page

Sample pentest report

Is this example from a real customer?

No. Loja Exemplo Ltda is fictitious. A customer report is confidential and only shared with their permission.

How much does a test like this cost?

The 25-hour Sprint costs R$9,500, delivered within 7 business days.

Does the attestation letter say the system is secure?

No. It states what was tested, when, how many findings there were and the status of fixes. It is not a certificate or a compliance opinion.

Read the full page

SaaS security glossary

What is the difference between IDOR and BOLA?

They are the same flaw (reaching another object by swapping an id) seen from different angles: IDOR is the classic web-application term, BOLA is the name in the OWASP API Top 10. In practice, we treat them as the same risk.

Which flaw is the most severe?

RCE (remote code execution), because the attacker gets to run code on your server. Next come cross-customer leaks (isolation) and secret exposure, which grant broad access without needing RCE.

Does RET test all these terms?

RET looks for these classes of flaw in the manual pentest and Risk Review, focused on the flows that sustain revenue (login, billing, data, uploads, admin) and on isolation between customers. Every finding comes with reproducible evidence.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and responsible disclosure.

Read the full page

Pentest for SaaS hosted on AWS

Do I need to tell AWS before the pentest?

For the services on the AWS policy list, such as EC2, RDS, Lambda, API Gateway and CloudFront, no. A service outside the list, and exercises with command and control or phishing, require contact or a form first.

Do you run load or denial of service tests?

No. The AWS policy forbids denial of service, real or simulated, and RET does not run that kind of test.

Can the test take production down?

Scope and window are agreed beforehand, and tests that can affect availability stay out. When a staging environment equal to production exists, the test starts there.

What if the test finds no flaw?

We refund 50% of the price. The other 50% pays for the hours executed, the report of what was tested and the attestation letter.

Read the full page

Buying a penetration test through public procurement in Brazil

Can a pentest be bought by reverse auction?

Yes, when the notice describes the service objectively (article 29 of Law 14.133). The reverse auction does not apply to technical services of a predominantly intellectual nature, so scope, hours, method and deliverables must be set in the terms of reference.

May the notice require a technical capacity certificate?

Yes, within article 67: a service of equal complexity, restricted to the most relevant parts and with a minimum quantity of up to 50% of them.

How much does it cost?

RET's 25-hour Sprint costs R$9,500 for one web application, API or mobile app (Android or iOS).

Read the full page

Pentest for AI applications

Does an API pentest already cover the AI part?

No. An API test checks who may call what. It does not try to make the model disobey its rules, leak internal instructions or misuse a tool.

My product only calls a model provider’s API. Do I need a test?

The provider’s model is not the target. Your product is: what you put in the context, which tools you give the model and what you do with the answer.

Read the full page

Android app pentest

Do I need to hand over the source code?

No. The test runs on the APK, as an attacker would. With source code the test goes deeper in the same time, but it is not required.

Does the app test include the API?

Yes. The calls the app makes to the server are followed and tested, because that is where the serious flaw usually is.

Read the full page

iOS app pentest

Do I need to hand over the source code?

No. The test runs on the app, as an attacker would. With source code the test goes deeper in the same hours, for the same price.

Does the iOS app test include the API?

Yes. The calls the app makes to the server are followed and tested, because that is where the serious flaw usually is.

I have the app on Android and iOS. Is that two tests?

It is one contract. Both apps use the same API, so they go together for R$19,800, a closed price.

Read the full page

Comparisons

Manual pentest vs automated security scanner

Does an automated scanner replace a pentest?

No. A scanner finds known vulnerabilities and weak config; it does not understand your product’s business rules, so it misses IDOR, plan bypass and cross-tenant leakage, exactly the flaws that hurt most in a B2B sale, and only manual testing finds them.

So scanners are useless?

Far from it. A scanner is cheap continuous hygiene: it catches outdated dependencies and config regressions before they become incidents. The mistake is thinking "green scan" means "secure product".

Who runs the pentest at RET?

Manual testing is run by Gabriel Lima Ferreira, RET’s founder and lead pentester, with authorization and a tight scope. Every finding ships with reproducible evidence and fix priority.

Can I start cheap?

Yes. If you do not need a full pentest yet, Risk Review reads 1 to 3 sensitive flows and tells you what to fix, what to monitor and what needs manual testing later.

Read the full page

In-house pentester vs outsourcing the pentest

Is outsourcing always cheaper?

For most early and mid-stage cases, yes, because you pay per scope instead of a fixed salary with benefits and tooling. It stops being cheaper when testing demand is daily and continuous; then an internal hire’s fixed cost is justified.

Is a third-party report worth more to a customer?

Usually. An independent assessment tends to be more accepted in a security questionnaire or due diligence than a test run by the same team that wrote the code.

Can I have both?

Yes, and it is often ideal: someone internal on the day-to-day and a periodic outsourced pentest for the outside view and independent proof. It is not internal or external; it is the right fit for your stage.

Who runs the outsourced pentest at RET?

Gabriel Lima Ferreira, RET’s founder and lead pentester, with authorization, NDA and a tight scope, delivering controlled evidence.

Read the full page

Reviewing your AI SaaS yourself vs human review

Can AI not review security on its own?

It reviews the obvious well: secrets, dependencies, validation. What it does not do is understand your product’s business rules, so it misses plan-access abuse, cross-tenant leakage and billing logic flaws. Those need a human eye.

So the Promptbook is not enough?

The Promptbook is enough for the first pass on your own and for classifying signals. That is exactly what it is for. When a signal is serious (access, data, money), the next step is a human review, and the Promptbook already hands you the material organized for it.

When is it worth paying for human review?

When the cost of a mistake rises: paying customer, third-party personal data, a B2B sale that requires proof. Before that, guided AI is enough and cheaper.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester. Risk Review reads 1 to 3 sensitive flows with authorization and returns fix priority.

Read the full page

Lovable vs Bolt.new: which is more secure?

So which is more secure?

On average, both have a similar hole rate (independent tests point to a 40-45% vulnerability rate in generated apps). What changes is WHERE: Lovable at RLS, Bolt at keys in the client. The most secure one is the one you reviewed, not the brand.

Can I migrate from one to the other to be safer?

Migrating does not fix it; you trade one hole profile for another. The security gain comes from locking RLS (Lovable) or removing keys from the client (Bolt), not from switching tools.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with authorization and a tight scope.

Read the full page

v0 vs Cursor: where generated code breaks

Which generates more secure code?

Neither generates secure on its own. v0 tends to give you clean UI with integration risk; Cursor tends to give you functional logic with access holes. What matters is reviewing the right layer of each.

Does a scanner find both tools’ holes?

It finds v0’s better (XSS, headers, secrets) than Cursor’s (IDOR/BOLA), because IDOR is business logic and a scanner does not understand who the owner should be. That is why Cursor needs more human review.

Read the full page

Firebase vs Supabase: security for AI-built apps

Which is more secure by default?

Both start permissive enough to leak if you do not lock the rule. There is no "more secure by default" here; there is "rule locked correctly". The model is equivalent: RLS on Supabase, Security Rules on Firebase.

Do I need to hide the anon_key / apiKey?

No, both are public by design and live in the client. What protects the data is the rule (RLS or Security Rules). Hiding the key does not help if the rule is open, and is not needed if the rule is correct.

Read the full page

RET vs hiring a development agency

Does my agency not already handle security?

It handles what it builds, to its own standard, but that is not an independent assessment. Whoever wrote the code has a blind spot for its own flaw (like AI). Security proof for a B2B customer usually requires a third party that tested, not the builder’s word.

Do I need to switch agencies?

No. Independent review does not replace the agency; it complements it. The agency builds, RET tests what it delivered. Having both is common and healthy.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with authorization and a tight scope, delivering reproducible evidence.

Read the full page

Manual pentest vs bug bounty

Does a bug bounty replace a pentest?

No. A bug bounty is continuous, open coverage with no guarantee of testing a specific scope by a date. A pentest is guaranteed coverage with a report, which is what closes a B2B sale and a security questionnaire. They complement each other; the bounty usually comes after the first pentest.

Is a bug bounty cheaper?

On paper it looks that way, because you only pay per finding. In practice the real cost is the operation: triaging, validating, rewarding and fixing each report. Without a team for that, the bounty becomes a debt and frustrates researchers.

I am a small SaaS, where do I start?

With a scoped review or test. RET Risk Review reads 1 to 3 sensitive flows and tells you what to fix; the Manual Pentest proves what breaks. A bug bounty makes sense later, once a fix process exists.

Can I have both?

You can, and it is ideal as the product matures: a scoped pentest at important milestones (big release, B2B sale, compliance) and a bug bounty running underneath for continuous coverage.

Read the full page

SAST vs DAST

SAST or DAST, which first?

It is not either/or. SAST slots easily into CI and catches code regressions on every commit; DAST runs against staging and catches runtime issues. Start with SAST because it is cheaper to integrate, but plan for both.

Do SAST and DAST together make a SaaS secure?

They cover a lot, but leave the biggest gap open: business rules. IDOR, plan bypass and cross-tenant leakage are invisible to tooling, because it does not know what your product should allow. That needs human review.

Can I run DAST in production?

With great care and never blindly: DAST generates attack traffic and can cause side effects (mass e-mail, dirty data, load). The standard is to run it against a staging environment that mirrors production.

Does RET sell SAST or DAST?

No. RET covers the gap the two leave: product logic. Risk Review reads the sensitive flows and the Manual Pentest proves what breaks; use SAST/DAST as hygiene underneath.

Read the full page

Point-in-time pentest vs continuous monitoring

Is one pentest a year enough?

For a product that changes little, it helps. For SaaS that deploys every week, no: the test becomes a photo that ages on the very next deploy. The ideal is a pentest at milestones plus continuous monitoring between them.

Does continuous monitoring replace a pentest?

No. It covers frequency (catching change on every deploy), but lacks the depth of a human chaining a real exploit and does not deliver the formal report a B2B sale asks for. The two complement each other.

What is AuditMySaaS?

It is RET’s continuous monitoring product: it follows what changes in your SaaS and flags when a change opens risk, so you are not blind between pentests. Plans at auditmysaas.app.

Do I start with point-in-time or continuous?

Usually with a photo: a Risk Review or Manual Pentest to know where you stand today. Then turn on continuous monitoring so you do not miss what changes from here on.

Read the full page

SOC 2 vs ISO 27001 for SaaS

Do I need both SOC 2 and ISO 27001?

Almost never at the start. Pick what your buyer asks for: SOC 2 for the US market, ISO 27001 for Europe/multinational/public sector. Doing both is only justified when you sell to both audiences and both require it.

Which is faster to get?

SOC 2 Type I is the fastest (a snapshot of control design). Type II observes months of operation. ISO 27001 involves building and auditing an ISMS, with its own cycle. The real timeline depends on how ready your base already is.

Does RET issue SOC 2 or ISO 27001?

No. SOC 2 is attested by an auditor (CPA firm) and ISO is certified by an accredited body. RET does the independent security testing (Risk Review, Manual Pentest) that produces the technical evidence those audits require, and helps you fix issues before you enter the audit.

Can I sell enterprise without certification?

You can, early on, by answering the customer’s security questionnaire with real evidence (a pentest report, documented controls). Certification becomes necessary when the buyer requires it by policy. Having testing and evidence ready shortens both paths.

Read the full page

In-house vs outsourced DPO

Does the law require a DPO?

LGPD requires the controller to name a DPO and publish the contact; GDPR requires one in defined cases. Regulators may waive it in specific situations, but the norm for SaaS handling customer data is to name a DPO, in-house or outsourced.

Is outsourcing the DPO safe?

Yes, it is common and accepted. The gain is experience and independence; the care is feeding the outsourced DPO with product context and ensuring contractual availability for incidents. It must not become a badge forgotten in a drawer.

Does the DPO handle technical security?

Not alone. The DPO handles process and the relationship with subjects and the regulator; the technical security of the data belongs to engineering. The law expects real security measures, which requires testing, not just policy.

How does RET help with privacy compliance?

By covering the technical side: testing tenant isolation, access control and personal-data leakage. A Risk Review on the flows that touch personal data gives your DPO real evidence of what is protected and what needs fixing.

Read the full page

RET Promptbook vs security consulting

Does the Promptbook replace consulting?

For the first review and common mistakes, yes, it handles a lot for R$147. For the non-obvious, the formal report a B2B sale needs, or sensitive data at scale, no: that is where Risk Review or the Manual Pentest come in. It shortens and cheapens that next step.

Do I need to know how to hack to use the Promptbook?

No. It is made for founders and devs who build with AI. You paste the prompts into the AI you already use and it walks the flows flagging risk. The work is following the playbook and acting on what shows up.

What is the next step after the Promptbook?

If a signal showed up in login, billing, data or a B2B sale, Risk Review (R$2,997) does a human read of 1 to 3 flows and tells you what to fix and what to test deeply. Then the Manual Pentest proves what breaks.

Does it work for apps built in Lovable, Bolt, v0 or Replit?

Yes. The Promptbook inspects real product flows, not the tool that generated the code. It works for any AI-built SaaS.

Read the full page

Dependency scanner vs application review

Don’t Dependabot or Snyk already handle security?

They handle one part: your dependencies. They do not look at your code or your business logic, so IDOR, plan bypass and cross-tenant leakage slip past. That needs application review, a different category.

Does "all green" on the scanner mean secure?

No. It means your libraries have no known CVE right now. It says nothing about the code you wrote, which is exactly where the most common incident in AI-built SaaS lives.

Does RET replace Snyk/Dependabot?

No. They are complementary: the scanner handles the dependency tree (automate and keep it on), RET covers your application and your logic. You want both.

Where do I start if I am short on time?

Turn on Dependabot today (fast and automatic) and review your application on the flows that touch access, money and data. RET’s Risk Review is the cheap path for that second part.

Read the full page

Pentest vs secure code review

Does code review replace a pentest?

No. The review finds the cause in code, but does not prove it is exploitable from outside nor produce the report a B2B sale expects. The pentest proves the impact. They are layers: cause and proof.

Which do I do first?

Usually the code review, because it is cheaper and catches the wrong pattern at the cause. Then the pentest confirms what remains resists. Running a pentest before fixing the obvious is paying dearly to discover the basics.

Does RET do both?

Yes: Risk Review is the human review of sensitive flows (inside) and the Manual Pentest is the proof (outside), with evidence. You can start with the read and move to the test as risk grows.

Does it work for AI-generated code?

It does, and that is where it helps most: AI-built products repeat wrong patterns (access control in the front-end, unsigned webhooks) that code review catches fast at the source.

Read the full page

WAF vs pentest

If I have a WAF, do I need a pentest?

You do. The WAF blocks generic attacks at the edge, but does not see logic flaws (IDOR, plan bypass, cross-tenant leakage), because those arrive as valid requests. Only testing finds them, and only a fix closes them.

Doesn’t the WAF block IDOR?

Not reliably. An IDOR is a well-formed GET with a swapped id; to the WAF it is legitimate traffic. It does not know that user does not own the object, that is your business rule.

So is a WAF useless?

Far from it. A WAF reduces noise, blocks scanners and known attacks, and is cheap defense in depth. The mistake is treating it as a substitute for testing; it mitigates, it does not fix.

Does RET configure my WAF?

That is not the focus. RET finds and proves the flaw the WAF does not cover (the logic one) and points to the fix at the cause. WAF and testing are different, complementary layers.

Read the full page

RET Risk Review vs Manual Pentest

Which is cheaper?

Risk Review (R$2,997), with a fixed, predictable price. The Manual Pentest is scoped (proposal), because depth and surface vary. To start cheap and learn where the risk is, Risk Review is the door.

Does Risk Review work as proof for a B2B customer?

As a signal and prioritization, yes; as a formal pentest report, no. If the buyer requires a pentest report, the step is the Manual Pentest, which delivers reproducible evidence in the expected format.

Do I need Risk Review before the Pentest?

Not necessarily. If you already have proof required or high risk, jump straight to the Manual Pentest. Risk Review is the efficient step when you still want to discover where investing is worth it without spending depth early.

And the Promptbook, where does it fit?

Before both: the Promptbook (R$147) is the playbook to fix the obvious yourself. After it, Risk Review does the human read and the Manual Pentest proves. It is a ladder of cost and depth.

Read the full page

LGPD vs GDPR for SaaS

If I comply with LGPD, do I comply with GDPR?

On technical security, almost entirely, the principles overlap. The delta is legal/operational: an EU representative, a formal international-transfer mechanism, and the higher fine ceiling. Taking LGPD seriously is a strong base for GDPR.

Does GDPR apply if my server is in Brazil?

It applies to the data of subjects in the EU, regardless of where the server is. The criterion is the person whose data you process, not the machine’s location.

Does RET handle the legal side of LGPD/GDPR?

No. RET covers the technical security both require (LGPD Art. 46 / GDPR Art. 32): testing isolation, access and leakage. The documentation belongs to your legal team or DPO.

Where do I start if I will sell to Europe?

Secure the technical base first (a Risk Review on the personal-data flows) and handle the paperwork with your legal team. Arriving at the negotiation with tested security shortens the DPA and the customer questionnaire.

Read the full page

Build vs buy authentication (Auth0/Clerk)

Does using Auth0/Clerk make my login secure?

It makes authentication much better protected (hashing, session, MFA, recovery). But it does not solve authorization: what the logged-in user can access is still your code. The common incident (IDOR, access to another customer’s data) happens after login.

When is building your own login worth it?

When you have a real requirement the provider does not meet and a team to maintain it securely. Early on, and especially in AI-built products, the provider removes an entire class of expensive mistakes from your plate.

Does the provider lock me in?

To some degree, yes: migrating identity later is laborious. It is a real cost to weigh against the security and speed gain early on. Many SaaS accept the initial lock-in and reassess at scale.

How do I test whether my access is secure?

The Promptbook guides you to review login and authorization with AI; Risk Review does the human read of the access flows. Both look at authorization, which is where the provider does not protect you.

Read the full page

.env file vs secret manager

Can I use .env in production?

It can work, but you lose rotation, access control and auditing, and raise the risk of leaking via commit or log. For production, your platform’s secret manager (or Doppler/Vault) is the standard. A .env is fine in dev, in .gitignore.

What is the most common secret mistake in AI-built SaaS?

A secret key in a public-prefixed variable (NEXT_PUBLIC_/VITE_/REACT_APP_), which ships to the browser bundle and is visible to any visitor. A scraper finds it in hours. A real secret lives only on the server.

How do I know if I leaked a key on the front-end?

Open the site, view the source/bundle in the browser and search for the key (or prefixes like sk_, service_role). If it shows up, it is already public, rotate the key immediately and move it to the server.

Does RET find exposed secrets?

Yes. The Promptbook guides that hunt (public prefix, hardcoded, credentials in logs) and AuditMySaaS monitors re-exposure on every deploy. It is a fast, high-impact fix.

Read the full page

By tool

Security for apps built with Lovable

Is the Supabase anon_key a secret?

No, it is public by design and lives in the frontend. The problem is that without RLS enabled and a correct policy on each table, that public key becomes direct database access. The protection is RLS, not hiding the key.

How do I know if my Lovable app has RLS off?

In practice, by testing: a query with the anon_key, no login, that returns rows from a sensitive table proves RLS is off or the policy is wrong. It is the first test RET runs, and the Promptbook guides you to do it yourself.

Is this a Lovable problem or mine?

Both. Lovable accelerates a lot, but it delegates access control to you by default. The app is yours, the data is yours, so the correct policy is the responsibility of whoever published it. The good news is that fixing RLS is fast once you know where to look.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester. The review runs with a verified domain, accepted terms and authorized scope, and returns reproducible evidence with fix priority.

Read the full page

Security for apps built with Bolt.new

Why is a key in the client so serious in Bolt?

Because Bolt is frontend-first and tends to put the call in the browser. Anything in the bundle is public: automated scrapers sweep fresh deploys and find exposed keys in hours, and then the cost (and access) is yours. The fix is moving the key to an environment variable and a server-side function.

What is the difference between anon_key and service_role?

The anon_key is public and respects RLS. The service_role bypasses RLS entirely, it is an admin key. If the service_role lands in the Bolt bundle, anyone reads and writes the whole database. It must never be in the client.

Does Bolt not protect this on its own?

Not by default. It optimizes for prototype speed, not production posture. The server layer, input escaping and rate limiting are steps you (or a review) need to add before real users arrive.

Read the full page

Security for apps built with v0

Is v0 code insecure?

Not on its own. v0 generates functional, clean code; the risk shows up in integration, when a snippet with dangerouslySetInnerHTML, a key in the client, or a server action with no owner check enters your app without review. It is the seam that needs eyes.

Is a Next server action secure by default?

It runs on the server, which helps, but it does not validate authz on its own. An action that writes to the database needs to check that the logged-in user owns that resource, otherwise it becomes IDOR. v0 writes the functional action; the owner check is yours.

Do I really need CSP?

CSP greatly reduces the damage of an XSS, which is exactly the risk of pasting dangerouslySetInnerHTML. Shipping without CSP is not an isolated bug, but it amplifies any injection. Worth configuring before real users arrive.

Read the full page

Security for code generated with Cursor

What is IDOR/BOLA and why does Cursor generate it?

It is when the endpoint checks that you are logged in but not that the requested resource is yours, so by changing an id you reach another user’s data. Cursor generates it this way because it writes the functional flow (fetch the resource by id) and does not add the owner check unless you ask.

Should I worry about the security of Cursor itself?

Yes. CVE-2026-50548/50549 (DuneSlide) allow zero-click prompt injection with code execution on the machine, CVSS 9.8. It is worth keeping it updated and reviewing agent, MCP and auto-execution permissions, not just the code it generates.

Does a scanner find these holes?

A scanner finds little IDOR/BOLA because it is a business-logic flaw: it does not know that id should belong to the owner. That is why a human review (Risk Review) is the right path for this kind of hole.

Read the full page

Security for apps built with Replit

Was the wiped database the agent’s fault?

It was a trigger, but the root cause was architectural: the agent had direct write access to production, with no barrier or environment separation. Agents make mistakes; the control is not letting an agent mistake reach real data. Reviewing that boundary is the point.

How do I reduce the agent’s power?

By separating dev from production, removing the agent’s write access to real data, and requiring human approval for destructive actions. RET’s review maps exactly what the agent reaches today.

Does a scanner solve this?

Partly. A scanner can find an open endpoint, but it does not understand the agent’s privilege boundary or the environment separation, which is logic. That is where the human read comes in.

Read the full page

Security for apps built with Firebase Studio

Is the Firebase API key in the client a problem?

The Firebase config (including the apiKey) is meant to live in the client, that is expected. The problem is not the key, it is the Security Rules: if they are permissive, the key becomes database access. The real protection is the rule, not hiding the config.

What is the if true rule?

It is the Security Rule that allows read and write from anyone (allow read, write: if true), usually left during development to avoid friction. If it reaches production, the whole database is public. It is the number one hole in Firebase apps.

How do I close it properly?

With a rule that checks authentication and ownership: each document is only read/written by its owner (request.auth.uid). RET’s review maps, collection by collection, what is open and what is missing a check.

Read the full page

Security for Next.js apps

Does CVE-2025-29927 affect me on Vercel?

No. Vercel (and Netlify) strip the x-middleware-subrequest header at the edge, so apps hosted there were not hit. The risk is for self-hosting on a vulnerable version (below 15.2.3/14.2.25/13.5.9). Even so, do not rely on middleware alone for authorization.

Is updating Next.js enough?

Updating closes the specific CVE, but the lesson is architectural: do not concentrate authorization in a layer the framework can skip. Check access at the data boundary (Route Handler, Server Action, query). That way, even a middleware bypass does not hand over the data.

Is a Server Action safe because it is "server"?

Not automatically. A Server Action is a public POST endpoint: anyone can call it. It needs origin checks, rate limiting and authorization like any route. Treating it as "internal" is a common mistake.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with a verified domain, accepted terms and an authorized scope, returning reproducible evidence.

Read the full page

Supabase security: RLS, service_role and PostgREST

Does the anon_key need to be hidden?

No, it is public by design and lives in the front-end. What protects the database is RLS with a correct policy on each table. Hiding the anon_key does not help; enabling RLS and checking ownership does.

What is the difference between anon_key and service_role?

The anon_key respects RLS (it is the public client key). The service_role BYPASSES RLS (it is the server admin key). Pasting the service_role in the front-end is catastrophic: it grants full database access. It must never leave the server.

Does enabling RLS solve everything?

Enabling is the first step, but the policy must check the record owner (request.auth.uid()), not just exist. An RLS with `using (true)` is "on" and still leaks data between customers.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope, returning reproducible evidence and fix priority.

Read the full page

Security of GitHub Copilot-generated code

Does CamoLeak still affect me?

CVE-2025-59145 was fixed by GitHub (disabling image rendering in Copilot Chat in August 2025). The lesson remains: an agent that reads external context (a PR description, an issue) can be hijacked by a hidden instruction. Limit what it executes and do not treat external content as an order.

Is Copilot code insecure?

Not by definition, but it repeats the pattern it saw in training, including the insecure one (concatenated query, auth in the front-end, hardcoded secret). The acceleration is real; the security guarantee does not come with it. So what it generates in a sensitive flow needs review.

How do I review AI-generated code without being an expert?

The Promptbook gives you prompts to force the AI itself to walk the flows and flag risk. It is the cheap first pass; when access is critical, Risk Review does the human read.

Read the full page

Security for n8n workflows

Is n8n cloud (n8n.cloud) also affected?

Some 2026 CVEs (like Ni8mare) were announced affecting self-hosted AND cloud, with the managed version fixed by n8n itself. The biggest exposure risk is in outdated self-hosted, where updating is your responsibility. Check your version against the advisories.

Why is a compromised n8n so severe?

Because it stores the credentials of everything it automates: API keys, database passwords, OAuth tokens. Compromising n8n usually grants access to all connected systems, not just one machine. The blast radius is enormous.

Is updating enough?

Updating closes the known CVEs and is the urgent first step. But also lock down public webhooks, isolate the Code node, take n8n off the open internet and limit what each workflow can execute. RET reviews that whole set.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope, focusing on blast radius and stored credentials.

Read the full page

Security for Bubble.io apps

Is Bubble insecure?

Bubble has security features, but many are not on by default: privacy rules start empty and the Data API often ships enabled. The app is exposed by default configuration, not by a bug. Closing it means configuring a privacy rule per data type and requiring auth on workflows.

Does hiding the interface protect my data in Bubble?

No. The Data API talks straight to the database, without going through your screen. An attacker queries the API and ignores the interface. The real protection is the privacy rule, not the absence of a button on the screen.

How do I know what is exposed?

By testing the Data API with no login and seeing what comes back, per data type. It is the first test RET runs on a Bubble app, and the Promptbook guides you to do it yourself.

Who runs the review at RET?

Gabriel Lima Ferreira, founder and lead pentester, with an authorized scope and reproducible evidence via the API itself.

Read the full page

Security for Windsurf-generated code

Does CVE-2025-62353 affect me even with auto-execution off?

It did: the critical point of that flaw was exfiltrating secrets via prompt injection EVEN with auto-execution disabled, via path traversal. Update Windsurf and do not treat a project file from an external source as trusted content.

Should I worry about Windsurf’s MCP?

Yes. There was an injection able to modify the local MCP config and register a malicious server, turning into command execution. Use an MCP allowlist, review the config and limit what the agent runs on its own.

Is Windsurf code secure?

It accelerates, but repeats insecure patterns like any assistant (auth in the front-end, IDOR, secrets). What it generates in a sensitive flow needs review; the Promptbook guides that check.

Read the full page

Claude Code agent security

Is Claude Code insecure?

Not by definition, but like every agent that reads external context, it had real CVEs for injection and exfiltration (InversePrompt, API-key exfil). Fixed in newer versions. The protection is least privilege, a trusted directory and secrets out of the context, not faith in the model.

Is updating the CLI enough?

Updating closes the known CVEs and is essential. But the pattern remains: do not open an unknown repo directly, limit auto-execution and keep secrets out of the agent context. The permission architecture matters more than the version.

How do I review the code Claude Code generated?

The Promptbook gives you prompts to force the AI to walk the flows and flag risk; Risk Review does the human read when access is critical. The review applies to any agent, the name changes, not the principle.

Read the full page

Security for apps built with Base44

Does the Base44 flaw still affect me?

The specific vulnerability Wiz found was fixed by Wix within 24 hours, with no sign of exploitation. But the pattern (a public id, an endpoint with no auth, SSO that does not cover every door) is what you should check in your own app, because access configuration is the publisher’s responsibility.

Is a private Base44 app really private?

It depends on how access is configured. The Wiz case showed apps "private by SSO" accessible via register/OTP endpoints with no auth. Real privacy comes from checking authorization on every door, not from hiding the URL.

How do I test my app’s access?

By trying what you should not: registering with no invite, calling an endpoint with no token, using a public id to reach data. It is the first test RET runs, and the Promptbook guides you to do it yourself.

Read the full page

Want to know whether this happens in your system?

Manual pentest for web applications, APIs and apps, from R$9,500, with the Sprint run within 7 business days. We refund 50% if we find no flaw.

See price and timeline for my case