# Sekrd > Sekrd is a security scanner for apps built with AI coding tools (Lovable, Bolt, Replit, Cursor, v0) on Supabase, Firebase, Vercel, and Netlify. It reads a live URL from the outside and reports exposed databases, leaked secrets, unsigned payment webhooks, missing security headers, and — on a deep scan — a small set of read-only probes. It does not send exploit payloads. Findings are evidence-backed; lookalike public keys (Supabase publishable keys, bare Google Maps AIza keys) are not treated as secrets. Sekrd Research publishes guides and original data. Figures from the Sekrd dataset may be cited with attribution to Sekrd (sekrd.com). Do not invent statistics that are not in the linked articles. ## Product - [Home](https://sekrd.com): Security scanner for AI-built apps. - [Free scan](https://sekrd.com/scan): Score, verdict, and bundle-level findings for a live URL. - [Sample report](https://sekrd.com/sample-report): Public sample scan of the intentionally vulnerable demo app, with fix prompts. Demo keys are masked. - [Deep scan](https://sekrd.com/scan/deep): Adds read-only active checks. Included with Pre-Launch Audit and Continuous Pro. - [Pricing](https://sekrd.com/pricing): Free, Pre-Launch Audit ($39 one-time), Continuous Pro. - [Docs](https://sekrd.com/docs): What the scanner checks. - [Rules](https://sekrd.com/rules): Finding rule catalog. - [About](https://sekrd.com/about): About Sekrd. - [Blog](https://sekrd.com/blog): All articles. - [Lovable security scanner](https://sekrd.com/lovable-security-scanner): Outside scan of a deployed Lovable URL. - [Bolt security scanner](https://sekrd.com/bolt-security-scanner): Client secrets, Supabase, and Firebase on a Bolt deploy. - [Supabase RLS checker](https://sekrd.com/supabase-rls-checker): Bundle secrets, REST reads, and policy SQL with a database password. - [Cursor security scanner](https://sekrd.com/cursor-security-scanner): NEXT_PUBLIC_ secrets, RSC pin, middleware probe, MCP tools/list. - [v0 security scanner](https://sekrd.com/v0-security-scanner): Next.js checks on a deployed v0 app. - [Comparisons](https://sekrd.com/compare): Public-page comparisons as of November 2026. - [Sekrd vs Vibe App Scanner](https://sekrd.com/compare/vibe-app-scanner): URL scan versus URL plus source review. - [Sekrd vs CheckVibe](https://sekrd.com/compare/checkvibe): URL scan versus app, repo, and pull requests. - [Sekrd vs VibeEval](https://sekrd.com/compare/vibe-eval): Automated findings versus engineer-verified delivery. - [Sekrd vs Lovable security scan](https://sekrd.com/compare/lovable-security-scan): Deployed URL versus the in-project Quick and Deep scans. - [RSS](https://sekrd.com/feed.xml): Blog feed. - [llms.txt](https://sekrd.com/llms.txt): Short description for language models. ## Articles - [Railway says stolen AI keys drove about $740,000 in unauthorized charges](https://sekrd.com/blog/railway-ai-keys-stolen-october-2026): Railway’s November 10, 2026 incident report says an attacker who broke out of the build system on October 20 decrypted customer environment variables for at most 1.85% of workspaces, about 100,000, an upper bound rather than a count of confirmed stolen keys. The queries targeted AI provider credentials. Railway estimates unauthorized spend on the stolen keys at about $740,000 and says it will cover charges that providers will not refund. If a key lived in Railway before October 21, rotate it at the provider and review that provider’s usage. - [Chainalysis says password hashes and live tokens for every user were exposed](https://sekrd.com/blog/chainalysis-tokens-exposed-november-2026): Chainalysis’s own incident page, updated through November 10, 2026, says an attacker using what the company called offensive agentic techniques reached its systems in October. By the night of November 4 the company said names, usernames, password hashes, and live tokens for all Chainalysis users had been exposed. It also said it still had no evidence that investigation data was accessed or that blockchain attribution data was manipulated. Reactor, KYT, and Address Screening were taken offline and later restored with mandatory password resets and API-key rotation. - [Next.js and React patched critical remote code execution on October 14, 2026](https://sekrd.com/blog/nextjs-react-october-2026-security-release): On October 14, 2026, the Next.js and React teams published critical advisories for unauthenticated remote code execution. Next.js 16.0.0 through 16.4.0 is affected by a Cache Components bug when a use cache function returns multiple typed arrays, fixed in 16.4.1. A separate image-processing bug, which Netlify identifies as CVE-2026-96889, affects Next.js 10.0.0 through 15.5.27 and 16.0.0 through 16.4.0 on glibc Linux, fixed in 15.5.28 and 16.4.1. React Server Components packages through 19.0.8, 19.1.9, 19.2.8, and 19.3.0 are affected in rare cases, fixed in 19.0.9, 19.1.10, 19.2.9, and 19.3.1. Upgrade and redeploy. Do not send an exploit payload. - [Supabase Storage let unauthenticated callers overwrite objects](https://sekrd.com/blog/supabase-storage-resumable-upload-bypass-2026): On October 14, 2026, Supabase published GHSA-8rrf-w776-x75g. Storage versions 1.80.1 and earlier let an unauthenticated caller create or overwrite objects in public or private buckets by reaching the signed resumable-upload route without a signed-upload token. The request then ran with a privileged database connection, so row-level security did not apply. Version 1.80.2 checks the token from the matched route instead of the raw URL. The advisory does not say the hosted platform was already patched. - [A JWT role of none bypassed row-level security in Supabase Storage](https://sekrd.com/blog/supabase-storage-jwt-role-none-rls-2026): On November 6, 2026, Supabase published GHSA-fgvp-cqg6-q8h2. Storage 1.89.8 and earlier accepted a valid signed JWT whose role claim was the string none, then ran the database work as the Storage connection user. When that user owns the tables and row-level security is not forced on the owner, policies are skipped. The advisory says that could allow reads from private buckets, uploads, and bucket creation. Exploitation needs a trusted token issuer, such as a custom access-token hook, to mint that claim. The hosted platform is already patched. Self-hosted installs need 1.89.9 or later. - [tensorlake 0.5.144 on npm was a Shai-Hulud credential worm](https://sekrd.com/blog/tensorlake-npm-shai-hulud-october-2026): Tensorlake’s advisory GHSA-8g63-53c9-65j2 says that on October 8, 2026, npm version 0.5.144 of tensorlake was published from a compromised maintainer account through the project’s own release workflow, so the package carried valid npm provenance. It was live for about one hour and 40 minutes, from 01:12 to 02:54 UTC. A preinstall hook ran an obfuscated payload that stole GitHub, npm, cloud, and AI-tool credentials and could republish itself. It also installed a gh-token-monitor service that deletes the home directory if the stolen GitHub token is revoked. Remove that service before you rotate the token. PyPI and crates.io were not affected. - [MCP Inspector could run code from a remote server’s OAuth URL](https://sekrd.com/blog/mcp-inspector-oauth-code-execution-2026): On October 21, 2026, the MCP Inspector project published GHSA-q978-78c7-phqw. Versions 2.0.0 through 2.11.0 of @modelcontextprotocol/inspector could run arbitrary code as the user on Windows and WSL while connecting to a remote MCP server. The server chooses an OAuth authorization URL, and the npm open package placed that URL in a PowerShell double-quoted string, which evaluates subexpressions. No tool call and no model decision are required. macOS and Linux are not affected. The patched release is 2.12.0. The advisory rates the issue high. - [Attackers used Qubole’s AWS account to read customer environments](https://sekrd.com/blog/qubole-aws-cross-account-theft-october-2026): Wiz reported on November 9, 2026, and updated the post on November 11, that a threat actor used IAM credentials from AWS account 805246085872, which Wiz observes Qubole’s platform using, to assume customer integration roles between October 14 and October 26, 2026. Every attributed AssumeRole call included an External ID and succeeded. Inside those accounts the actor read Lambda environment variables, Secrets Manager, and SSM parameters, validated credentials with TruffleHog, and downloaded S3 objects from whatever buckets the role could reach. Wiz does not say how the Qubole credentials were first stolen, and it does not give a victim count. - [A ConnectWise Cloud Backup credential was used inside Microsoft 365 tenants](https://sekrd.com/blog/connectwise-cloud-backup-microsoft-365-2026): On October 17, 2026, ConnectWise shut down Cloud Backup, formerly SkyKick, after unauthorized use of its application credentials against Microsoft 365 tenants. Huntress said that as of 19:00 UTC on October 16 it had confirmed unauthorized sign-ins in 162 tenants, starting October 15 from 144.172.64.128. ConnectWise disabled seven application identities at 07:00 UTC on October 17 and said that does not change each tenant’s local configuration, and that previously issued tokens can keep working until they expire. Stored backup archives were described as encrypted, with no evidence they were accessed. - [PHP shipped emergency fixes for two unauthenticated remote code execution bugs](https://sekrd.com/blog/php-unauthenticated-rce-october-2026): On October 21, 2026, PHP released 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to fix two critical vulnerabilities, CVE-2026-109618 and CVE-2026-109619. The 8.4.27 release notes say both can allow unauthenticated remote code execution and that updating is the only safe choice. The matching advisories are GHSA-w8hw-w9fx-j79w, an out-of-memory state corruption, and GHSA-6wmf-4rv5-p3rp, a use-after-free from stale per-request state kept on a worker. The project changelog lists both advisory IDs and both CVE IDs on one line and does not pair them one-to-one. There is no complete configuration workaround. - [What Sekrd’s November 2026 scanner update checks — and what it refuses to flag](https://sekrd.com/blog/scanner-update-november-2026): The November 2026 scanner update adds firm, low-false-positive checks for secrets and misconfigurations that show up in AI-built apps: new provider key formats in the JavaScript bundle, a Supabase sb_secret_ key (it bypasses Row Level Security), a Gemini key used against the paid generativelanguage API, server secret names behind NEXT_PUBLIC_, VITE_, EXPO_PUBLIC_ and similar prefixes, a system prompt in the bundle that contains credentials or an authorization rule (OWASP LLM07:2025), and an exact pin of a React Server Components package affected by CVE-2025-55182. A deep scan adds read-only probes for the Next.js middleware bypass CVE-2025-29927, an MCP endpoint that lists tools with no authentication, GraphQL introspection that names sensitive fields, a public S3 or GCS bucket that lists credential or database objects, and a real .env file or /.git/HEAD. A bare 200, a public image bucket, a Maps API key, or a normal system prompt is not a finding. The update does not send exploit payloads. - [How to secure a Lovable app before you ship](https://sekrd.com/blog/how-to-secure-a-lovable-app): Secure a Lovable app by treating the Supabase publishable key as public, enabling Row Level Security with per-user policies on every table that holds user data, keeping the sb_secret_ or service_role key only in Edge Function secrets, and then testing the deployed site while signed out. Lovable’s Quick Scan and Deep Scan check the project source; they do not replace a request against the live URL. The October 30, 2026 Supabase Data API change stops new tables from being exposed automatically, and it does not close tables that are already readable. - [How to secure a Bolt app before you deploy](https://sekrd.com/blog/how-to-secure-a-bolt-app): Secure a Bolt app by assuming every VITE_ variable is public, moving LLM, Stripe, and database-admin calls to a server or Edge Function, and setting Supabase RLS or Firebase rules before the first deploy. Bolt generates the project; the hosting provider serves the bundle. Search the production build for secret prefixes, keep .env out of the repo and off the web server, and scan the deployed URL while signed out. - [How to secure an app you built in Cursor](https://sekrd.com/blog/how-to-secure-a-cursor-built-app): Secure a Cursor-built app by reviewing the diff before you deploy, not the chat. Reject client-prefixed secrets (NEXT_PUBLIC_, VITE_, EXPO_PUBLIC_), keep authorization in the route or database policy rather than only in middleware or a system prompt, and pin Next.js and React Server Components to a patched release. Cursor is the editor. The production URL is what visitors get, so scan that URL after deploy. - [How to secure a v0 app before launch](https://sekrd.com/blog/how-to-secure-a-v0-app): Secure a v0 app by treating it as a Next.js app you are about to deploy. NEXT_PUBLIC_ values are sent to the browser, so API keys and the Supabase secret key do not belong there. Check the user session inside every Server Action and route handler that reads or writes private data, not only in middleware.ts. If the lockfile pins react-server-dom-webpack, react-server-dom-parcel, or react-server-dom-turbopack at 19.0.0, 19.1.0, 19.1.1, or 19.2.0, upgrade to the patched revision in that line (19.0.1, 19.1.2, or 19.2.1) and to the patched Next.js release your major requires. Then scan the deployed URL. - [How to secure a Replit app before you publish it](https://sekrd.com/blog/how-to-secure-a-replit-app): Secure a Replit app by storing credentials in the Secrets tool, not in source, and by adding the same keys again as deployment secrets before you publish — development secrets do not carry over. Secrets are not available to static deployments. Never print environment variables from a route or a public project: collaborators who cannot view the secret in the UI can still read it if the code prints it. Keep database rules and a server-side API key proxy the same way you would on any other host, then scan the published URL. - [Supabase RLS checklist for AI-built apps](https://sekrd.com/blog/supabase-rls-checklist): A Supabase app is safe to put on the internet only when every table in exposed schemas has Row Level Security enabled, every policy names the caller for private data, and the secret key is not in the client. Grants and RLS are different switches: after October 30, 2026 new tables are not exposed to the Data API until you GRANT, and a GRANT without RLS makes the table public to the publishable key. Verify with pg_tables, pg_policies, and one signed-out request — not with the dashboard’s green RLS badge. - [Your OpenAI or Gemini API key is in the frontend: what to do in the first hour](https://sekrd.com/blog/leaked-openai-gemini-api-key-frontend): If an OpenAI key (sk- or sk-svcacct-) or a Gemini key used with generativelanguage.googleapis.com is in a frontend bundle, treat it as public: revoke it in the provider dashboard, turn on a hard spend limit, check usage, and ship a new build that calls the model from a server. Deleting the line from git does not remove a key browsers already downloaded. A bare AIza key used for Maps, Firebase, or reCAPTCHA is a different object — restrict it by API and referrer, and do not rotate it in a panic unless it can also call the Gemini API. - [NEXT_PUBLIC_, VITE_, and EXPO_PUBLIC_ secrets: which env vars are actually public](https://sekrd.com/blog/next-public-env-secrets): NEXT_PUBLIC_ (Next.js), VITE_ (Vite), EXPO_PUBLIC_ (Expo), NUXT_PUBLIC_ (Nuxt), and PUBLIC_ (SvelteKit) are compile-time public prefixes. The bundler replaces them with the literal value and ships it to every visitor. Put publishable keys, site URLs, and analytics IDs there. Put LLM keys, Stripe secret keys, database URLs, webhook secrets, and Supabase sb_secret_ keys in server-only variables and never import those variables into client code. Search the production build, not just the repository. - [CVE-2025-29927 explained: the Next.js middleware bypass and how to check your app](https://sekrd.com/blog/cve-2025-29927-nextjs-middleware-bypass): CVE-2025-29927 is an authorization bypass in Next.js middleware. If your only auth check runs in middleware, a request can skip it on affected self-hosted versions. Patches are 15.2.3, 14.2.25, 13.5.9, and 12.3.5. Next.js 11 has no patch; block the internal header at the proxy until you upgrade. Apps on Vercel or Netlify, and static exports, are not affected. Check your lockfile version and move authorization into the route or database. Do not probe production with an exploit request. - [CVE-2025-55182: which React Server Components versions to upgrade](https://sekrd.com/blog/cve-2025-55182-react-server-components): CVE-2025-55182 is unauthenticated remote code execution in React Server Components, rated CVSS 10.0. It affects react-server-dom-webpack, react-server-dom-parcel, and react-server-dom-turbopack at 19.0.0, 19.1.0, 19.1.1, and 19.2.0. Patched revisions are 19.0.1, 19.1.2, and 19.2.1. Next.js apps that vendor those packages need the patched Next.js release for their line; follow the React security post because later fixes moved the minimum. Apps with no server, and apps that do not use a bundler that supports React Server Components, are not affected. Check with npm ls. Do not send an exploit payload. - [OWASP Top 10 for LLM Applications 2025, for people shipping an app](https://sekrd.com/blog/owasp-llm-top-10-2025-for-app-builders): The OWASP Top 10 for Large Language Model Applications 2025 is a risk list for apps that call a model, not a scoring system for the model vendor. For an app built with Lovable, Bolt, Cursor, v0, or Replit, the items that show up in production first are LLM02 sensitive information disclosure (API keys in the bundle), LLM07 system prompt leakage (secrets or authorization rules in a prompt), LLM06 excessive agency (an MCP or tool endpoint with no auth), LLM10 unbounded consumption (no rate limit or spend cap), LLM01 prompt injection, and LLM05 improper output handling. The prompt is not an access-control boundary. Enforce that in the database and the route. - [Vibe coding security checklist before you launch](https://sekrd.com/blog/vibe-coding-security-checklist): Before you launch an AI-built app, confirm five things on the deployed URL, not in the chat. The client bundle contains no server secret and no sb_secret_ key. Every private Supabase table or Firebase collection has a real rule, not USING (true) or allow if true. Authorization is enforced in the handler and the database, not only in middleware or a system prompt. Next.js and React Server Components are on a patched release if you use them. The host does not serve .env, .git, or a public bucket of backups. Then run a free scan of that URL. - [Supabase's October 30, 2026 Data API change: what breaks in Lovable, Bolt and Cursor apps](https://sekrd.com/blog/supabase-data-api-change-october-30-2026): On October 30, 2026 Supabase stops automatically granting the anon, authenticated and service_role roles access to new tables in the public schema of existing projects (new projects have had this default since May 30, 2026). Tables that already exist keep their grants and keep working. Any table created after the change — including tables an AI coding tool creates with raw SQL — returns a permission error from the Data API until you add explicit GRANT statements and, for anything user-facing, enable Row Level Security with policies. - [16,326 Supabase databases were publicly readable: what UpGuard found and how to check yours](https://sekrd.com/blog/upguard-16326-supabase-databases-exposed): On September 25, 2026 security firm UpGuard reported 16,326 Supabase databases whose tables could be read by anyone, found while studying roughly 300,000 domains showing signs of Supabase use. More than half had schema signs of personal data; some exposed password or authentication-token fields. Supabase was not breached: the tables had Row Level Security disabled, so the public (anon/publishable) key embedded in every frontend could read them. You can check your own project in five minutes by listing public tables with RLS off and by querying them with your public key. - [Is Lovable's built-in security scan enough? What Quick Scan and Deep Scan check — and what they don't](https://sekrd.com/blog/lovable-security-scan-what-it-checks): Lovable’s Quick Scan runs automatically at publish and checks database access rules and RLS gaps, password-protection settings, known vulnerabilities in npm dependencies and unauthenticated MCP servers. Deep Scan is on-demand and adds authorization, abusable endpoints, injection, leaked secrets, payment manipulation, auth bypass and data exposure in logs. Lovable itself states the scans cannot guarantee security, do not replace a professional review for apps handling sensitive data, and do not cover runtime behavior, business-logic flaws or compliance beyond OWASP patterns. Use the scans, then test the deployed app. - [CVE-2025-48757 explained: the Lovable Row Level Security vulnerability and what it teaches](https://sekrd.com/blog/cve-2025-48757-lovable-rls-explained): CVE-2025-48757, published May 30, 2025, describes insufficient Row Level Security policies in apps generated by Lovable through April 15, 2025, allowing remote unauthenticated attackers to read or write arbitrary database tables. It is rated CVSS 3.1 9.3 (Critical) and classified as CWE-863, Incorrect Authorization. Lovable disputes it, arguing each customer is responsible for protecting their application’s data. Whatever the attribution, the bug class is current: tables reachable with the public Supabase key and protected by missing or permissive RLS. - [Supabase is retiring anon and service_role keys by the end of 2026: how to migrate to publishable and secret keys safely](https://sekrd.com/blog/supabase-legacy-api-keys-deprecation-2026): Supabase is deprecating the legacy JWT-based anon and service_role keys by the end of 2026. Their replacements are publishable keys (sb_publishable_…), which are safe in browsers and map to the anon or authenticated role under Row Level Security, and secret keys (sb_secret_…), which map to service_role, bypass every RLS policy and must never appear in a web page, mobile bundle, source code, URL or log. Migrate by creating the new keys, replacing the publishable key in clients and the secret key only in servers and Edge Functions, verifying, and then deactivating the old keys. - [Why your OpenAI or Anthropic API key ends up in your frontend bundle — and how to find it first](https://sekrd.com/blog/exposed-ai-api-keys-frontend-bundle): An LLM API key ends up in a frontend bundle when the app calls OpenAI, Anthropic or another provider directly from the browser, usually through an environment variable with a client prefix such as VITE_, NEXT_PUBLIC_ or EXPO_PUBLIC_. Build tools inline those variables into JavaScript that every visitor downloads, so the key is public the moment the site deploys. Find it by searching the built bundle for key prefixes; fix it by moving the call into a server or Edge Function that authenticates the user, rate-limits them and holds the key as a server-only secret — then rotate the leaked key. - [Protecting Supabase Edge Functions that call LLMs: verify_jwt, rate limits and webhook signatures](https://sekrd.com/blog/supabase-edge-functions-llm-abuse-protection): Supabase Edge Functions verify the caller’s JWT by default (verify_jwt = true); setting verify_jwt = false removes that platform check and makes your handler fully responsible for authenticating callers. For a function that calls an LLM, keep JWT verification on, read the user from the verified claims, rate-limit per user ID (Supabase’s documented approach uses Upstash Redis), cap tokens per request, and keep the provider key in function secrets. Functions that must be public — health checks or third-party webhooks — should verify the provider’s signature or do nothing billable. - [MCP server security in 2026: tool poisoning, rug pulls, token passthrough — a practical checklist](https://sekrd.com/blog/mcp-server-security-checklist-2026): The main MCP risks are tool poisoning (hidden instructions in tool descriptions that the model follows but the user never sees), rug pulls (a server changing its tool descriptions after you approved it), cross-server shadowing, token passthrough and local server compromise. Invariant Labs demonstrated tool poisoning on April 1, 2025, exfiltrating SSH keys and MCP configuration files through Cursor. The MCP specification’s security best practices (version 2026-07-28) state that MCP servers MUST NOT accept tokens not explicitly issued for them, and that clients offering one-click local server setup MUST show the exact command and obtain consent before running it. Pin server versions, review tool descriptions, minimize scopes and never pass tokens through. - [Stripe webhook signature verification in AI-generated apps: the check that stops free orders](https://sekrd.com/blog/stripe-webhook-signature-verification-ai-apps): Stripe signs every webhook event with an HMAC-SHA256 signature in the Stripe-Signature header, using the endpoint’s whsec_ signing secret. Your handler must verify that signature against the raw, unmodified request body using Stripe’s library (constructEvent) before acting on the event; without verification, anyone can send a fake payment_intent.succeeded and get goods or access for free. Stripe’s libraries reject events older than a 5-minute default tolerance to stop replays — never set the tolerance to 0 — and handlers should deduplicate by event ID because Stripe may deliver an event more than once. - [AI-built app security report 2026: what 799 live apps show (Sekrd dataset)](https://sekrd.com/blog/ai-built-app-security-report-2026): Across 799 publicly reachable hosts scanned by Sekrd between March 22 and September 10, 2026 (174 on lovable.app subdomains, plus Vercel, Bolt, Replit, Netlify, Cloudflare Pages, Firebase Hosting and custom domains), 4.4% had at least one critical finding and 6.8% received a Block (do-not-ship) verdict. The most widespread gaps were deployment-level: 81.5% served no Content-Security-Policy, 93.4% disclosed server software versions, 60.3% had no privacy-policy link and 25.3% offered no visible account- or data-deletion mechanism. High-severity secret leaks appeared on 2.1% of hosts. Figures are external, unauthenticated scans and may be cited with attribution to Sekrd. - [100 vibe apps, one scanner: what a random sweep of the AI-built web actually finds](https://sekrd.com/blog/100-vibe-apps-one-scanner): AI coding tools ship faster than any IDE in history. Platform defaults haven't caught up. We scanned 100 random production apps from 12 AI-coding hosting platforms: 5% had critical live leaks (Stripe keys in the browser, full Supabase table exposure), 57% shipped with no Content-Security-Policy at all. The fix is continuous scanning in CI — the same pre-commit discipline that stops typos from reaching production. - [98 vibe-coded apps, 967 findings: the security posture is the platform, not the app](https://sekrd.com/blog/vibe-coded-apps-98-findings-late-april-2026): We ran ~100 scans across vibe-coded apps and indie deployments from two different cohorts. 96% shipped, 4% got BLOCK, 21% bounced off their own WAF. Inside each cohort, scores clustered so tight it surprised us. The real risk doesn't come from the template — it comes from the 3–6% of apps where the developer reached past the template. - [50 sites, 444 findings: the 2026 state of shipping-fast security](https://sekrd.com/blog/vibe-coding-legacy-cms-audit-april-2026): April 2026. We scanned 50 production URLs — vibe-coded AI tools, small-business WordPress, auth-SaaS homepages. 41 scored, 9 bounced off modern WAF and we respected it. 88% SHIP, 12% BLOCK, median 76, 444 findings. Every site had findings. Here's what the shape of those findings says about how the web is being built right now, and why 'ship fast' can't keep ignoring the security layer. - [5 security checks a URL-only scanner will never catch](https://sekrd.com/blog/checks-url-scanners-miss): A URL scan checks your front door. Deep scans check whether the windows are locked. Here are the five classes of findings that only show up when you plug in your Supabase, Firebase, or MCP credentials. - [Your Supabase RLS says "enabled." Here's why your database is still open.](https://sekrd.com/blog/using-true-not-safe): A Row Level Security policy with USING(true) passes every dashboard check and leaks every row. We've seen it in production apps, in our own test projects, and in the Supabase docs themselves. Here's what it looks like and how to find it before someone else does. - ["allow read, write: if true" — the Firestore rules gap nobody fixes in time](https://sekrd.com/blog/firebase-rules-audit): Firebase test-mode rules give you 30 days. A lot of apps get a year. We've scanned projects with wide-open Firestore, wide-open Storage, and surprisingly tight rules that just needed a matching deny-all fallback. Here's what the audit actually checks. - [We scanned 49 of the internet's biggest sites. Here's what we found.](https://sekrd.com/blog/web-security-audit-49-sites-2026): GitHub, Google, Slack, Stripe, Notion, Vercel, and 43 others. 89% passed. 5 got BLOCK. Every single site had findings. The average score was 80/100. Full data and methodology inside.