I find vulnerabilities in the applications of companies like Apple, Google, Microsoft,
Goldman Sachs, SpaceX and GitHub — most of it inside private, invitation-only bug bounty programs.
Six years on HackerOne: top-1,000 all-time, HackerOne Clear, 131 paid bounties.
I work AI-augmented: my own harnesses, built around Claude and Codex, run reconnaissance,
triage and reporting at a scale one hunter couldn't reach by hand.
Open to penetration testing, red teaming and vulnerability research roles · remote worldwide or relocating · GMT+6
Reputation, rank, signal and impact are checkable on my public HackerOne profile;
report, severity and bounty counts come from my HackerOne account.
0Reputation
0Global rank top 1,000 all-time
0Reports resolved + 80 triaged
0Critical & high findings 27 critical
0Programs that thanked me of 182 reported to
0Paid bounties 36 programs
0Signal 80th percentile
0Impact top 16% for severity
HackerOne ClearBackground-checked. Most researchers are not.
9-month streakValid reports every month, Jan – Sep 2026.
Milestone Level IIIAll three tiers cleared, 2025–26 cycle.
18 badgesSix of them distinct OWASP categories.
02 Highlights
The findings I can point to.
Most of my work is under private-program terms and can't be written up. These are the results
that are on the record — dated, and linked to the source where one exists.
Two bounties from Goldman Sachs.3 May and 24 April 2026 — the second co-authored with fellow Bangladeshi hunter refat0g.
Ranked #1 of every hunter on a private HackerOne program.Forty-eight valid reports and 1,075 reputation on a single program. It can't be named, as its terms require — it is the best evidence of what I do.
Apple security acknowledgements, twice in two months.July 2025 and August 2025.
Acknowledged by UNESCO.A reflected cross-site scripting flaw on unesco.org resources, reported 19 June 2025 — listed by name on UNESCO's disclosure page.
Public hall of fame at SpaceX, Atlassian, Twilio and the U.S. Department of Veterans Affairs.Sixty-four valid findings on a second platform; three more programs private.
Joined HackerOne.Everything above happened in the six years since.
03 Every wave
The organisations that put their name to my findings.
Every mark below is a program that accepted at least one of my reports as valid — the public
ones, on both platforms — or an organisation that lists me by name on its own acknowledgements
page. Rendered in a single ink, so no logo shouts over another — and every mark links to the program's own page.
And these are only the ones I'm allowed to name. Most of my work is in private programs whose names are withheld, as their terms require.
04 What I find
Ten OWASP Top 10 classes, from injection to SSRF.
Not a skills list. Six carry HackerOne's own badge — issued only on a valid report of that class; the other four are on my record as resolved or triaged reports.
BadgeInjectionSQL and command injection — untrusted input reaching an interpreter.
BadgeBroken AuthenticationSession and identity flaws — the class that ends in account takeover.
BadgeXML External EntitiesParsers coaxed into reading files and reaching internal services.
BadgeBroken Access ControlIDOR and privilege escalation — the object is there, the check is not.
BadgeSecurity MisconfigurationDefaults left in place, surfaces left open, secrets left reachable.
BadgeCross-Site ScriptingStored, reflected and DOM-based execution in another user's session.
ValidSensitive Data ExposureInformation disclosure — my most frequent class: PII, tokens and internals reachable without authorisation.
ValidServer-Side Request ForgeryThe server made to fetch on my behalf — into metadata endpoints and internal services.
ValidOpen RedirectTrust in a link, redirected — the first step of many phishing and token-theft chains.
ValidInsecure DesignBusiness-logic flaws — flows that work as built and fail as intended.
Where some of my resolved-or-triaged reports landed, in HackerOne's own weakness classes:
45Information disclosure
37Access control
31Authentication
22Cross-site scripting
14Open redirect
6SSRF
4Insecure design
3Injection
05 Tools
Reconnaissance at this scale needs tools nobody ships.
Extracts hosts from Shodan and defeats the platform's hard ceiling of 1 000 results
per query by recursively subdividing the search — country, then city, then port — until
every bucket fits under the limit. Validates, strips reserved ranges, deduplicates.
A port scanner that never touches the target. Resolves open ports out of Shodan's
InternetDB at ~500 requests a second with rotating user agents, so an entire estate can
be mapped without a single packet reaching the asset owner.
AI-augmented pentesting
Claude · Codex · personal harnesses
I run offensive work through harnesses I built around Claude and Codex — agents wired to
my own reconnaissance tooling that map an estate, triage what comes back, draft proofs of
concept and write the report, with me steering every exploitation step. It is how one hunter
keeps up with an attack surface that never stops growing.
Claude
Codex
Agent harnesses
Tool use
Recon automation
Triage
Reporting
Attack Surface Monitoring
SaaS · in development
Continuous external attack-surface mapping, folding Crunchbase, Shodan, Censys,
Whoxy and crt.sh into one picture with custom reconnaissance on top.
Rust
Python
Bash
Shodan
Censys
crt.sh
Whoxy
06 Languages
Four languages.
Fluent in English and Bengali; conversational Hindi and Urdu — speaking and listening.