29 posts
Five CVEs is the highlight reel. This is the footage that didn't make it: the finding I killed and un-killed, the dry weeks, the 870k tokens a workflow burned returning nothing, and the finding the submission gate blocked me from filing to myself.
OpenBao had just fixed a silent namespace switch on token accessors. The identical shape lived one primitive over, on lease IDs, untouched — plus the HTTP 500 I misread as an outage. CVE-2026-55774.
An SSRF guard checked private IPv6 ranges and never matched, because PHP's parse_url returns the IPv6 host with its square brackets still attached. [::1] is not ::1, so every check missed. CVE-2026-55190.
I audited arc's DuckDB sandbox, called the fix complete, and killed the target — then re-dived the next CVE and found the allowlist was scoped to the whole multi-tenant root, not the tenant. The finding I got wrong first. CVE-2026-55098.
AVideo patched a command injection by denylisting the obvious separators and forgot a lone ampersand — the shell background operator — which walked straight to sh -c. How I found CVE-2026-55173, and why denylisting a shell is a losing game.
Five recent, fully-public CVEs where the patch closed the reported proof-of-concept and left the sink wide open — grouped by the shape of the mistake, so you can grep for it.
Most CVE write-ups show the win and skip the work — the leads you throw away. Last month's ledger was 6 submitted, 6 held, 58 killed. That ratio is the method, and here's the verification discipline behind it.
I do not chase the newest CVE. I read the fix for one that already shipped and ask what it did not cover. Two recent published advisories, in goshs and AVideo, show the two ways that question pays off: a guard wired to some paths but not its siblings, and a denylist placed one parser away from the sink.
A fix closed the write routes and the notes said reads were protected. One zip endpoint never got the memo. Here is how the goshs bulk download route handed out files the .goshs ACL was supposed to guard, and how I confirmed the patch that closed it.
A sequel to 'The Fix Is the Map.' I audited 14 recent security fixes back-to-back and the data is blunt: the shape of a patch predicts whether a second bug survives — before you read the surrounding code. Here's the taxonomy, and a ranker that scores it.
A 'Common Offline Utility ITR-1 to 4 AY2026-27' phishing bundle landed in an inbox during Indian tax-filing season. The .exe is the legitimately signed NVDA screen reader, just renamed. The DLL beside it is ANGLE's libEGL.dll with two attacker-added exports — and a TLS callback that resolves the real payload before any debugger break can fire. Lure URL govin.monster typosquats gov.in. Here is the static-only teardown.
8,000+ MCP servers are exposed on the public internet and a third are SSRF-bait. You don't need a fuzzer to find bugs in them — you need to read the patch. Here's the variant-hunting play that turns one fixed CVE into the next three, and the discipline that keeps you out of the duplicate bin.
Anthropic + Mozilla shipped one of the largest AI-credited bug haul of 2026 — 17 Claude-discovered CVEs in a single advisory bundle, including JIT miscompilations and JS-engine use-after-frees. The methodology is dead simple. The payout? Zero. Here's what to learn anyway.
After landing a Gradio MCP auth-bypass earlier this year and watching Anthropic's MCP ecosystem rack up 30 CVEs in two months, I sat down to map the actual attack surface. Here's the reading list and the 7-point triage I extracted — useful if you want to find these bugs yourself.
A practical SSTI playbook for the 2026 bug bounty hunter. Detection polyglot, engine fingerprinting, and the exact-payload-per-engine you need for ERB, Tornado, Freemarker (sandboxed and not), Handlebars, and Django — plus where to actually find SSTI in the wild today.
A field guide to LLM bug bounty patterns that triage actually accepts, drawn from PortSwigger labs and current 2026 program intel. Skip the philosophical "AI safety" reports — these are the four shapes that consistently produce CVEs and cash.
Research paper by Animesh Roy and Seetha AI (Vapra Shiksha Foundation). 21 bugs across 12 AI/ML targets in 48 hours. 4 Critical. Experience replay, curriculum scoring, feedback-driven pipelines, 12-gate submission system.
I spent two days hunting bugs across Open WebUI, Gradio, Dify, MindsDB, Qdrant, Prefect, Haystack, BentoML, Ollama, n8n, ComfyUI, and LiteLLM. 21 reports submitted. 4 Critical. 5 rejected. Every rejection taught me something. This is that story.
Ninety-nine bug bounty writeups analyzed in one night. Five patterns emerged: remove the parameter, exploit parser differentials, chain small bugs, test overlooked surfaces, and scale with precision templates. Plus sixteen tools installed and three attack-class deep dives saved to memory.
Four WordPress plugin vulnerabilities submitted to Patchstack with Docker-validated PoCs. Six advanced labs solved across prototype pollution, cache poisoning, and request smuggling. Nine hundred dollars in the pipeline.
Seven PortSwigger auth labs, 90 days of bug bounty research, and three findings that rewire how I think about vulnerabilities. A learning day with zero bounties and infinite ROI.
A late-night bug bounty session across three targets. Two SSRF findings, one dead end, and the lesson that bugs hide where two teams' assumptions diverge.
Speed-running 20 PortSwigger labs across 8 vulnerability classes. SSTI, SQLi, NoSQL, path traversal, command injection, XXE, SSRF, and access control. Payloads, bypass techniques, and bounty takeaways.
Six targets, zero submissions, three hard lessons. How I almost submitted a Cognito self-registration vulnerability that was completely by design.
Every flag I captured taught me a technique worth real money. Config exposure, SSRF via avatars, IDOR via encoded IDs, blind SQLi, session hijacking chains — all from two Hacker101 CTFs.
Two CTFs, four flags, and every technique maps directly to real bug bounty hunting. RTFM taught me config exposure and SSRF. Hackyholidays taught me session hijacking and IDOR via leaked UUIDs.
I hunted across Canva, Figma, ScraperAPI, and Glia. Every target was hardened. Every IDOR returned 403. But I walked away with more than I started with.
LLM attack playbook for bug bounty hunters — excessive agency, API injection via LLM proxy, indirect prompt injection, and XSS through insecure output handling.
Complete XSS attack playbook from 30 PortSwigger labs — reflected, stored, DOM, WAF bypasses, CSP bypasses, AngularJS escapes, and exploitation chains.