# Excession Labs > Excession Labs is a security research lab. It runs authorized security assessments of web apps, APIs, multi-tenant SaaS, LLM apps and AI agents for businesses, and publishes reproducible security research. Key facts: - Services: security assessment of a web app, API or SaaS; security audit of one product, with each weak point ranked by risk on four severity levels; AI assurance for LLM apps, agents and MCP servers; recurring tests after releases; retests of fixed findings; a free pilot for a small scope. - Method: the engagement splits into tasks that AI testing agents run in parallel, as each user role in scope. Each tool runs, waits for approval or is blocked, by rules set for the scope. A finding counts only with evidence: the request, the response and the controls that held. Excession Labs checks the evidence for each finding and writes the report. - Report: HTML, PDF and Markdown, with proof-of-concept scripts. Each finding has evidence, reproduction steps and a fix. Controls that held are listed too. - Authorization: testing starts only after written authorization from the system owner. Do not send passwords or keys. - Prices: quoted after scoping. No prices are published. - Contact: the form at https://excessionlabs.ai/#request, or support@excessionlabs.ai. Each page below is Markdown. All text in one file: https://excessionlabs.ai/llms-full.txt ## For agents - CVE intelligence (public, read-only Streamable HTTP): https://excessionlabs.ai/mcp/cve. Tools: cve_search, cve_get, cve_history, cve_status, cve_sources and cve_changes. Search more than 424,000 CVEs with software/version evidence, CISA KEV and dated FIRST EPSS scores. Check source freshness and pagination; software matches do not establish applicability. These six tools are also available at the main MCP endpoint. - MCP server (Streamable HTTP): https://excessionlabs.ai/mcp. Read-only tools without a key: search_site, get_page, how_to_contact. The tool request_security_test sends a security test request for a person who asked for one; it needs an agent key from POST https://excessionlabs.ai/agent/auth, sent as Authorization: Bearer . The person must agree to be contacted and must own or be allowed to arrange a test of the systems. Limits: 5 requests a day per key, 3 a day per email address. - Server card: https://excessionlabs.ai/.well-known/mcp/server-card.json - OpenAPI (agent key and security test request): https://excessionlabs.ai/openapi.json - Agent authentication: https://excessionlabs.ai/auth.md ## Lab - [CVE intelligence](https://excessionlabs.ai/cve-intelligence.md): connector setup, tools, source coverage, version evidence and collection limits - [Home](https://excessionlabs.ai/index.md): services, method, report contents, request form - [Application security assessment](https://excessionlabs.ai/security-assessment.md): what we assess, test methods, evidence rules, safety limits, report contents - [Penetration testing services](https://excessionlabs.ai/penetration-testing.md): authorized penetration tests of web apps, APIs, SaaS and AI apps; scope, process, report, quote - [Web application penetration testing](https://excessionlabs.ai/web-application-penetration-testing.md): sign-in, sessions, access control, injection and browser-side tests - [AI penetration testing](https://excessionlabs.ai/ai-penetration-testing.md): AI testing agents under per-scope rules, with people who check each finding; tests of AI apps - [External penetration testing](https://excessionlabs.ai/external-penetration-testing.md): tests of internet-facing web apps, APIs, admin panels and login flows from outside - [AI red teaming for LLM apps and agents](https://excessionlabs.ai/ai-red-teaming.md): prompt injection, tool abuse and data exposure in LLM apps, agents and MCP servers - [API penetration testing](https://excessionlabs.ai/api-penetration-testing.md): authorization, mass assignment, injection and limits in REST, GraphQL and other HTTP APIs - [Sample penetration test report](https://excessionlabs.ai/sample-report.md): a sample report for a fictional app, with the PDF at https://excessionlabs.ai/sample-report/excession-labs-sample-report.pdf - [Pricing](https://excessionlabs.ai/pricing.md): no price list; what sets a written quote, what it includes, the free pilot - [About Excession Labs](https://excessionlabs.ai/about.md): the lab, how the work is split, authorization, contact ## Research - [How to Prepare for a Web Application Security Audit](https://excessionlabs.ai/blog/prepare-web-application-security-audit.md): how to prepare for a web application security audit: scope and written approval, test accounts, synthetic data, evidence handling, monitoring and recovery, and a readiness checklist. - [How often must you pentest? What PCI DSS, HIPAA, ISO 27001 and SOC 2 say](https://excessionlabs.ai/blog/how-often-penetration-testing.md): pCI DSS sets a 12-month pentest interval. HIPAA, ISO 27001 and SOC 2 set none in the text we could read. Clause numbers, quotes, trigger events and the date we read each source. - [AI Red Teaming: How to Plan a Controlled Assessment](https://excessionlabs.ai/blog/ai-red-teaming-plan-controlled-assessment.md): how to plan a controlled AI red teaming assessment: scope and approval, test cases for prompt injection, tool permissions and cross-tenant leaks, containment, stop rules, evidence and retests. - [MCP security: a vulnerable MCP server you can run, with three flaws and their fixes](https://excessionlabs.ai/blog/mcp-security.md): a local MCP server with a vulnerable mode and a fixed mode, and a test for each of three flaws: hidden instructions in a tool description, token passthrough and missing per-user authorization. All 4 attack checks succeed in vulnerable mode and all 4 are blocked in fixed mode. Quotes the 2026-07-28 MCP specification. - [SaaS Penetration Testing: 8 Checks for Tenant Isolation](https://excessionlabs.ai/blog/saas-penetration-testing-tenant-isolation-checks.md): eight checks for SaaS tenant isolation: tenant context, objects, search, roles, bulk actions, files, background jobs and caches, with what a pass and a fail looks like for each. - [Does SOC 2 require a penetration test? What the criteria say](https://excessionlabs.ai/blog/soc-2-penetration-testing.md): no: the SOC 2 Trust Services Criteria do not require a penetration test. CC4.1 names one as an example of an evaluation, and two CPA firms say it is not required. We quote the criteria, list the evidence to keep and cover test timing for Type I and Type II reports. - [API Security Testing: 7 Checks for Access Control Gaps](https://excessionlabs.ai/blog/api-security-access-control-checks.md): seven tests that expose API access control gaps: missing credentials, cross-user and cross-tenant access, role and field permissions, and alternate routes. - [Vulnerability assessment vs penetration testing: 4 of 27 flaws found](https://excessionlabs.ai/blog/vulnerability-scanner-vs-penetration-test.md): we ran OWASP ZAP and Nuclei against 27 known flaws in local test apps. The best of six setups found 3 and missed 23. All six together found 4 and found 2 more in part. 21 flaws were missed by every setup, including SQL injection, five login-flow flaws and six authorization flaws. - [Prompt injection defenses: success rates from 1,800 trials](https://excessionlabs.ai/blog/prompt-injection-defenses.md): 1,800 lab trials against a mock support agent: with no defense, 69 of 300 prompt injection attacks succeeded. Spotlighting cut that to 28, code checks to 33, simulated human approval to 36 and a URL allowlist to 48. No defense stopped every attack. - [Penetration testing cost in 2026: 68 published prices](https://excessionlabs.ai/blog/penetration-testing-cost.md): of 66 pentest providers checked on 9 November 2026, 21 publish 68 price rows. For US-dollar prices sold per test, the median is $5,000 across 21 rows from 9 providers, from $1,000 to $27,120. 14 of the 21 rows are starting prices. - [ReDoS: catastrophic backtracking in a Python regex](https://excessionlabs.ai/blog/redos.md): a nested-quantifier validation regex took about 3.1 seconds on a 27-character input in a local Python endpoint, and its time quadrupled with every two extra letters. A linear pattern with a length cap answered the same input in 6.6 ms. - [HSTS: test the Strict-Transport-Security header locally](https://excessionlabs.ai/blog/hsts.md): we ran four HSTS header setups against a local server with curl: no header, a short max-age, a header sent only over plain HTTP, and a correct one. Only the correct setup came back with no findings. - [PKCE: testing authorization code interception locally](https://excessionlabs.ai/blog/pkce.md): a local toy OAuth server issued tokens to a party holding only an intercepted authorization code. Requiring an S256 code challenge denied those redemptions with invalid_grant, while the legitimate client still got its token. - [Credential stuffing: a local test and the fixes that work](https://excessionlabs.ai/blog/credential-stuffing.md): we replayed 100 invented leaked username and password pairs against a local login. The per-account lockout never fired and 12 of 60 synthetic accounts opened, then a per-client attempt budget, a leaked-password check and a second factor were each retested with real recorded responses. - [HttpOnly cookie: a local test of cookie flags](https://excessionlabs.ai/blog/httponly-cookie.md): a session cookie set without HttpOnly, Secure or SameSite and with a wide Domain failed a local audit with five problems. We record both headers with curl, model the rules in Python, and fix it with a __Host- cookie. - [Zip slip: reproduce archive path traversal locally](https://excessionlabs.ai/blog/zip-slip.md): a local upload-and-extract endpoint shows zip slip with ../ names, absolute names and a tar symlink member, how Python's zipfile and tarfile handle the same archives, and a resolve-and-refuse check that passes the retest. - [OWASP API Top 10: tests you can run, and a checklist](https://excessionlabs.ai/blog/owasp-api-top-10.md): each OWASP API Security Top 10 (2023) risk with one test your team can run, plus a local fixture that runs an API security checklist of 12 requests. - [IDOR vulnerability: a local example with PUT and DELETE](https://excessionlabs.ai/blog/idor-vulnerability.md): an IDOR vulnerability example: a second user lists, edits and deletes another user's invoices. UUIDs do not fix it; a server-side ownership check does. - [SQL injection example: login bypass and UNION reads](https://excessionlabs.ai/blog/sql-injection.md): reproduce SQL injection in a local sqlite3 API: login bypass, a UNION read of another table, the parameterized fix, and an ORDER BY gap needing an allowlist. - [SSRF: reproduce, bypass, and fix a preview endpoint](https://excessionlabs.ai/blog/ssrf.md): we reproduce an SSRF flaw in a URL-preview endpoint, bypass a naive hostname block with a redirect and an IP encoding, then fix and retest it. - [Timing attack: an API token check that returns early](https://excessionlabs.ai/blog/timing-attack.md): measure a timing attack on an API token check: compare == and compare_digest, recover token bytes over HTTP with a stated delay, then retest the fix. - [Clickjacking attack: a local frame-ancestors test](https://excessionlabs.ai/blog/clickjacking.md): reproduce a clickjacking attack on a local page, then block it with frame-ancestors and X-Frame-Options. Real curl headers and Chromium results included. - [Row level security: three Postgres mistakes, tested](https://excessionlabs.ai/blog/row-level-security.md): postgres row level security fails silently in three common ways. We reproduced each with real psql output on a throwaway cluster, then fixed and retested. - [security.txt: RFC 9116 rules, example and checker](https://excessionlabs.ai/blog/security-txt.md): learn what security.txt needs under RFC 9116, copy a valid example, and run a Python checker on five broken files and the corrected version. - [Path traversal: a local directory traversal experiment](https://excessionlabs.ai/blog/path-traversal.md): reproduce a path traversal attack on a local file-download fixture with curl, watch a strip-the-dots fix fail, then retest a real-path check. - [Session fixation attack: a local curl experiment](https://excessionlabs.ai/blog/session-fixation.md): reproduce session fixation in a localhost fixture: a planted session ID stays valid after login. Read 12 real responses, apply the fix and retest the old ID. - [API rate limiting: how to test a login limit](https://excessionlabs.ai/blog/api-rate-limiting.md): test API rate limiting on a local login endpoint: see the 429, bypass it by rotating X-Forwarded-For, then key on account and peer address and retest. - [Broken access control: vertical privilege escalation](https://excessionlabs.ai/blog/broken-access-control.md): reproduce three broken access control flaws in a local fixture: a hidden admin route, a client-set role header and a DELETE gap. Then fix and retest. - [GraphQL introspection: disabling it leaves the field open](https://excessionlabs.ai/blog/graphql-introspection.md): reproduce an unlisted GraphQL admin field locally, show it stays callable after introspection is off, then fix the resolver and retest 24 requests. - [2FA bypass testing for your own login flow](https://excessionlabs.ai/blog/2fa-bypass.md): reproduce three two-factor login flaws in a local fixture: step skipping, no code attempt limit, and a code not bound to one session. Fix and retest each. - [CSRF vulnerability: a local SameSite cookie experiment](https://excessionlabs.ai/blog/csrf-samesite.md): reproduce a CSRF vulnerability in a local change-email endpoint, record what Chromium sends for each SameSite cookie, then fix it with a token and Origin check. - [Open redirect vulnerability: testing a ?next= login fix](https://excessionlabs.ai/blog/open-redirect.md): reproduce an open redirect on a login ?next= parameter, defeat two naive fixes with //host and /\host values, then retest a parsed same-site path check. - [Mass assignment vulnerability: a local reproduction](https://excessionlabs.ai/blog/mass-assignment.md): reproduce a mass assignment vulnerability in a localhost profile API: a user sets role and credit, reads a hash stand-in, then we fix and retest. - [Tenant isolation testing: three gaps beyond IDOR](https://excessionlabs.ai/blog/tenant-isolation.md): reproduce three tenant isolation gaps in a localhost fixture: a client tenant_id filter, a shared cache key and an unchecked export job. Fix and retest. - [Race condition vulnerability: a reproducible coupon test](https://excessionlabs.ai/blog/race-condition.md): we reproduced a race condition vulnerability in a coupon endpoint, counted redemptions across repeated runs, and closed it with one atomic SQL update. - [Host header injection: a reproducible password-reset test](https://excessionlabs.ai/blog/host-header-injection.md): reproduce host header injection in a local password-reset fixture, inspect real HTTP responses, enable the canonical-origin fix, and retest the boundary. - [Authorization testing for APIs: a reproducible local workflow](https://excessionlabs.ai/blog/api-authorization-testing.md): reproduce a cross-account read on a localhost fixture and verify a server-side fix with recorded curl evidence. ## Policies - [Acceptable Use Policy](https://excessionlabs.ai/aup.md) - [Terms](https://excessionlabs.ai/terms.md) - [Privacy](https://excessionlabs.ai/privacy.md)