Korrosiv.AI
Dissolve your attack surface._
An AI-native penetration testing engine for web applications and APIs. Point it at a target. Watch it think, adapt, and find what others miss.
Dissolve your attack surface._
An AI-native penetration testing engine for web applications and APIs. Point it at a target. Watch it think, adapt, and find what others miss.
Korrosiv.AI — Client Portal
Your engagements, findings, and sealed reports — presented exactly as your team sees them.
· https://juice-shop.acme.test
· https://payments.acme.test
· Started 04 May 2026
9f2c4ab77e13d05aBrowser surface mapping
Access-control probing
Auth and session handling
Reproduced and confirmed findings
| Severity | Title | Category | Targets | Status |
|---|---|---|---|---|
| Critical | Cross-User Basket Access via IDOR | Access Control | https://juice-shop.acme.test/rest/basket/{id} | Open |
| Critical | Authentication Bypass via SQL Injection on Login | Injection | https://juice-shop.acme.test/rest/user/login | Open |
| High | Stored Script Executes in Customer Feedback | Injection | https://juice-shop.acme.test/#/complain | Open |
| High | Administration Console Reachable without Role Check | Access Control | https://juice-shop.acme.test/#/administration | Open |
| High | JWT Accepted with alg Set to None | Authentication | https://juice-shop.acme.test/rest/user/whoami | Open |
The GET /rest/basket/{id} endpoint authorises on the caller’s session but never checks that the basket in the path belongs to that session. Basket identifiers are sequential integers, so any authenticated customer can walk the range and read the contents of every other customer’s basket, including the products, quantities and the delivery address attached to the order. The same missing check applies to the write path, so a basket can also be modified by a caller who does not own it.
These reproduction notes are presented separately from exact tool-boundary captures.
Authenticate as a normal customer and read your own basket to establish the baseline
Reviewed requestcurl -s -X GET 'https://juice-shop.acme.test/rest/basket/2' -H 'Authorization: Bearer <user-1>' -w '\nHTTP_STATUS:%{http_code}'"id":2,"UserId":2 and the caller reads the basket that belongs to themExpected outcome: 200 with UserId matching the session, confirming default server behaviour
Request a basket belonging to a different customer using the same session token
Reviewed requestcurl -s -X GET 'https://juice-shop.acme.test/rest/basket/3' -H 'Authorization: Bearer <user-1>' -w '\nHTTP_STATUS:%{http_code}'"id":3,"UserId":3 along with another customer’s products, so no ownership check is appliedExpected outcome: 401 or 403; the server instead returned another customer’s basket verbatim
Confirm the identifier space is enumerable and the write path shares the flaw
Reviewed requestfor i in $(seq 1 12); do curl -s -o /dev/null -w "%{http_code} " \
"https://juice-shop.acme.test/rest/basket/$i" -H 'Authorization: Bearer <user-1>'; donePUT /rest/basket/3 was also accepted, so the record can be altered as well as readExpected outcome: non-owned identifiers rejected; every identifier in the range was readable and writable
Independently validated
Any authenticated customer can read and modify every other customer’s basket by incrementing a single integer in the path. Twelve of twelve identifiers tested returned another account’s data, and the write path accepted the same unowned identifier, so the flaw is not limited to disclosure.
Order contents and delivery addresses for the entire customer base are readable from a single free account, and orders can be altered before fulfilment. That is a reportable personal-data exposure as well as a direct route to fraudulent orders.
01Every engagement, one surface.
Client portal, in active development. Sample organisation and findings — not a customer engagement.
The walkthrough needs a wider screen to be legible.
Traditional penetration testing is a point-in-time exercise. You engage a team, wait weeks for availability, then receive a report that reflects the application as it existed during a narrow testing window.
Meanwhile, your development team ships daily. New features, new endpoints, new attack surface, all deployed between the time a finding is documented and the time it's read. The report is outdated before the ink dries.
Manual human-led pentests aren't scaling to the speed of modern development. The vulnerability lifecycle has compressed from months to hours, but the testing model hasn't changed in a decade.
Pentest engagement begins
Manual testing in progress
Report delivered
Dev team shipped 47 deployments, 12 new API endpoints, and 3 major features, all untested
The average web application receives multiple updates per week. A quarterly pentest covers less than 2% of the changes shipped.
Korrosiv.AI is an AI-native, AI-first automated penetration testing engine.
It does what a human pentester does, but without the time constraints, fatigue, or blind spots. It thinks through application logic, chains vulnerabilities, and adapts its approach in real time.
Built by pentesters who understood that the future of offensive security isn't more tools, it's smarter ones.
Every attack vector is methodical and targeted. We don't brute force, we dissolve.
Built by pentesters, for pentesters. The platform respects its audience's expertise.
Corrosion is not chaos, it's chemistry. Every test is deliberate and documented.
Real vulnerabilities. Real methodology. See how Korrosiv.AI reasons through complex attack chains that human pentesters miss under time pressure.
Starting from a single 404, Korrosiv.AI autonomously mapped 9 backend microservices, extracted every API route from exposed metrics endpoints, and uncovered an unauthenticated IDOR exposing ~72,000 citizen records with full PII. 12 minutes, zero credentials required.
Korrosiv.AI reverse-engineered a production JavaScript bundle, traced lazy-loaded webpack chunks, resolved chunk hashes, and pivoted through secured MFA endpoints to discover an unauthenticated password reset flow leaking plaintext security answers. 94 autonomous reasoning steps to full account takeover.
Full-stack analysis of web applications, testing authentication flows, session management, and probing business logic the way a human pentester would.
Automated discovery and exploitation of REST and GraphQL endpoints. Broken auth, injection, mass assignment, tested systematically.
The AI is the pentester. It reads responses, adapts payloads, and chains findings the way a senior consultant would, but reasons on 100% of the data.
Every finding includes reproduction steps, evidence screenshots, impact analysis, and remediation guidance your team can act on immediately.
Watch the AI work through a live dashboard. See endpoints discovered, tests executed, and vulnerabilities found, as they happen.
Built with guardrails. Scoped testing, controlled exploitation, and full audit trails. Offensive capability with defensive discipline.
On an internal infrastructure engagement, a spreadsheet of legacy credentials turned up on a file share. One password kept recurring, a string that looked completely random. No dictionary match, no known pattern. A human pentester moves on.
AI didn't. It identified the string as a keyboard walk pattern, recommended adding it to the active password spray, and that single insight corroded through layers of vendor defences into full domain compromise.
In a typical human-led pentest, AI reasons on roughly 5-20% of the engagement context. The rest gets skimmed or missed under time pressure. If that number reaches 100%, the outcomes change completely.
From September 2026 we take on a limited number of engagements. We scope it, we run the engine against it, and we validate every finding. You get a live reporting dashboard and a report your engineers can act on.
We only test what you own or are authorised to test. Scope and written authorisation are agreed before an engagement starts.
Korrosiv.AI — Engagements
Cohort 01 · first engagements September 2026
No spam. We’ll reach out to scope, and nothing else.