#title: "Bug Bounty as an Income Stream" subtitle: "Platforms, programs, specializations, and what hunters actually earn" audience: "Kris Yotam (Linux/C/Python/Rust, OSCP track, math background, AI-security interest)" related: "./pentest.md (cert + career-path side)" last_updated: 2026-05-20

#Bug Bounty as an Income Stream

This is the platforms-and-programs side. The cert track, formal pentesting employment, OSCP/OSWE/OSEP planning, and Synack Red Team vetting live in ./pentest.md. Read that first if you want the career-path angle; this document assumes you already know what XSS and a reverse shell are.

#1. Major BB Platforms — Head-to-Head

Ranking for a serious researcher in 2026:

1. HackerOne broadest program list. OpenAI is not here but Anthropic, GitHub, PayPal, Shopify, DoD, GitLab are. Best LHE pipeline.
2. Immunefi if you can do Solidity audits at all, the expected hourly here beats every other platform combined. The catch is the skill ramp.
3. Bugcrowd for OpenAI, Tesla, Atlassian, and roughly half the Fortune 500 private programs.
4. Intigriti fastest-growing, best triage, and EU public-sector programs you cannot get elsewhere.
5. Code4rena / Sherlock / Cantina pick one as a "compete on weekends" platform once you have Foundry skills.
6. YesWeHack fine as a secondary, especially if you want Asian/EU enterprise scope.
7. Synack covered in pentest.md; it is hybrid contractor/bounty work, not pure BB.

#2. Top-Paying Public Programs (2025-2026 Snapshot)

Time-to-pay reputation (broadly): GitHub, Shopify, Stripe, GitLab fast. Apple, Google, Microsoft slow but reliable. Tesla, Uber variable. Immunefi top programs pay within 2-8 weeks of fix deploy; KYC required above roughly $10k.

#3. AI / LLM-Specific Bug Bounties

This is your declared niche, so it gets a real treatment. The most important framing: classical AppSec bugs in AI products pay far better than novel alignment attacks. A prompt injection that exfiltrates another user's chat history via the plugin sandbox is an IDOR + SSRF story dressed in LLM clothing, and it pays IDOR + SSRF money, not "interesting paper" money.

#Programs

#Status paths (non-monetary but career-forming)

• HackAPrompt (2023, recurring) Learn Prompting + AI Village competition. Established the prompt-injection field; winners get hired.
• AI Village at DEF CON generative red-team competitions (the 2023 White House GRT, the 2024 follow-ups). Speaking here is a credential.
• OWASP LLM Top 10 the canonical taxonomy. LLM01 prompt injection through LLM10 supply chain. Use the IDs in your reports; triagers know them.
• MITRE ATLAS adversarial ML taxonomy. Useful for framing model-extraction / data-poisoning reports.

#Where the easy wins are in AI products (in 2026)

1. Indirect prompt injection that lands a real impact agentic systems (Operator, Atlas, Gemini-in-Workspace, Copilot Studio) that read attacker-controlled content and then act with the user's privileges. This is the single hottest 2026 category.
2. RAG data leaks system prompts, embedding-DB content, other users' chunks. Treat the vector DB as a data store and look for IDOR / auth.
3. Plugin / tool / MCP server escapes SSRF and RCE through a function-calling tool that takes URLs or shell args. Classic AppSec, AI surface.
4. Auth and sharing bugs in chat/file resources IDOR on share links, conversation IDs, file IDs, project IDs.
5. Sandboxing escapes code interpreter / "Advanced Data Analysis"-style containers.
6. Billing / abuse free-tier bypass, refund logic, account spam, model-cost evasion.

Your math + Python + Rust profile is unusually well suited to the agent / tool-use surface. That is the lane.

#4. Specializations and Their Economics

The AI/LLM line above is the income-realistic take. The "win the universal-jailbreak prize" line is closer to a contest payout than a steady stream.

#5. Realistic Income Data

Public sources: HackerOne's Hacker-Powered Security Report (7th edition, 2023-2024 data), Bugcrowd's Inside the Mind of a Hacker reports, and Immunefi's annual web3 bug bounty stats.

Median for active hunters is widely cited as roughly $0. Most reports submitted are duplicates, informational, or N/A. Among hunters who get even one bug paid in a year, the median is closer to $1k-$3k. The mean is dragged up by a tiny number of full-timers.

Going full-time before you have roughly $80k in trailing-12-month earnings is a budget-killer. The income variance is the real story: 6-month droughts between $40k criticals are routine.

Public earners worth following (their writeups, not their dollar figures, are what matters):

• @brutelogic XSS legend
• @nahamsec plus his streams
• @infosec_au (Shubs) recon at scale
• @cdl assetnote / now Searchlight Cyber
• @samwcyo Sam Curry, automotive and API bugs
• @stokfredrik methodology streamer
• @samczsun top-tier smart-contract auditor; writeups are textbook material
• @inversecos cloud forensics
• @orange_8361 top-tier server-side and proxy bugs

#6. Workflow and Tools

Brief, since you live in Linux. Treat the BlackArch laptop as the bounty rig; keep BB recon out of ~/dev/ and into a separate ~/recon/ or ~/ops/<program>/ tree with strict per-program directories so scope mistakes cannot happen.

#Recon stack

• amass, subfinder, assetfinder subdomain enum
• httpx alive probe + fingerprint
• nuclei templated scanning (essential)
• katana, gospider crawling
• gau, waybackurls historical URL discovery
• ffuf content + parameter fuzzing
• dnsx, dalfox, kxss DNS / XSS pipeline pieces

#Proxies

• Burp Suite Professional $475/yr, still the industry standard. Required for any program that uses session replay.
• Caido fast-growing Rust-based competitor with much nicer UX; free tier covers most work. Pair with Burp; do not replace yet.

#Smart contracts

• Foundry toolchain (forge / cast / anvil / chisel). Default in 2026.
• Slither static analysis
• Mythril, Echidna, Halmos symbolic / fuzz / SMT analysis
• Heimdall decompiler, Rust-friendly

#Mobile

• Frida, objection, jadx Android
• MobSF automated triage
• iOS: jailbroken device or Corellium virtualisation

#Cloud

• ScoutSuite, Prowler multi-cloud posture
• Pacu AWS exploitation
• ROADtools, TokenTactics Entra / M365

#AI-product testing

• Garak, PyRIT, promptfoo LLM red-team frameworks
• Burp AI extensions request tampering for LLM endpoints
• Custom Python harnesses for API replay; you already have the Python skills

#Pipelines

• Axiom distribute recon to spot VMs
• Trickest visual recon workflows, generous free tier

#7. Reporting and Comms

Quality dominates quantity. A high-signal hunter sends fewer than five reports a week; a low-signal hunter sends fifty. Triagers remember both.

A great report has:

1. One-line summary (severity + asset + bug class)
2. CVSS v3.1 vector (or the platform's own VRT / severity mapping)
3. Steps to reproduce minimal, copy-pasteable, no marketing
4. Proof of concept curl-able, or a short video; strip credentials
5. Impact what an attacker actually achieves, with realistic preconditions
6. Recommended fix short, does not say "use a WAF"
7. References OWASP, CWE, your own prior writeups

Severity disputes: programs use CVSS but most weight business impact. Lead with impact and the score follows. Bugcrowd uses its VRT; learn the categories before submitting.

Triage SLAs (approx):

• HackerOne managed: 1-3 business days first response
• Bugcrowd managed: 1-5 business days
• Intigriti: 1-2 days
• Immunefi: 1-7 days (a Slack/Discord war room appears for criticals)
• Self-managed programs: anywhere from hours to weeks

Duplicates are the dominant negative outcome for new hunters. Two mitigations: pick low-traffic asset classes, and report fast (within hours of discovery).

Disclosure-agreement risk is real: some programs change scope or rewards retroactively. Screenshot the policy on submission and save the rules / scope page.

#8. Reputation and Status Building

• Leaderboards hackerone.com/leaderboard, Bugcrowd's MVP rankings, Immunefi's Hall of Fame. Top-100 is realistic in 2-3 years for a strong hunter. Top-10 is a full-time job.
• Live Hacking Events (LHEs) H1's flagship invite-only on-site events (often in Las Vegas, Amsterdam, Tokyo) with six-figure event purses, two-day hack windows, plus the only place you will meet the other hunters. Bugcrowd runs equivalents. Earning an LHE invite is the first major status step.
• Public disclosure every report H1/Bugcrowd resolves can be requested for public disclosure. A polished writeup on your own site (krisyotam.com slot looks ideal) becomes recruiting bait.
• Twitter/X + Bluesky + Mastodon the BB community is still mostly on X. Following and posting concise findings is how new programs notice you.
• Speaking DEF CON AI Village, BSides anywhere, LASCON, Hexacon, Insomni'hack, OffensiveCon. Talks get you onto private programs.

#9. Tax and Business Setup (US)

Cross-link: see ./pentest.md business-entity section for the longer treatment shared with consulting income.

• BB income is 1099-NEC (or 1099-MISC for older platforms). PayPal/Coinbase payouts still generate 1099-Ks at $600+.
• Self-employment tax = 15.3% on top of income tax (Social Security 12.4% to the cap + Medicare 2.9%).
• LLC pass-through is fine until roughly $60-100k of profit; at that point an S-corp election (form 2553) lets you split into reasonable salary + distributions and save roughly 7-8% in SE tax. Get a CPA; the worst-case downside is IRS reclassification.
• Quarterly estimated taxes due Apr 15, Jun 15, Sep 15, Jan 15. Use IRS Direct Pay.
• Payment methods by platform: HackerOne (PayPal, Coinbase, Hyperwallet bank, Payoneer), Bugcrowd (PayPal, Payoneer, US bank), Intigriti (SEPA, Wise), YesWeHack (SEPA, bank wire), Immunefi (USDC on-chain or bank, KYC above thresholds), Synack (1099 + bank). Crypto payouts have their own basis-tracking burden; log every payout date and USD value at receipt.

#10. Risk and Legal

• DOJ 2022 policy update on CFAA press release good-faith research is no longer prosecuted under CFAA. This is a policy, not law; state laws still apply.
• Always verify scope on the day you test. Programs edit scope quietly.
• DMCA section 1201 still bites on circumvention of access controls in research that is not covered by the security-research exemption (renewed every three years). Mobile / firmware reverse-engineering lives here; check the current exemption text.
• Out-of-scope discoveries if you find something on a target's infra that is not covered, the right play is coordinated disclosure through their security.txt or the CISA CVD process, not a bounty submission and not a tweet.
• Never test production without explicit authorization. A security.txt pointing to HackerOne is not authorization for adjacent infra. Read every program's "Targets in Scope" list and stop at the boundary.
• Safe Harbor language in program rules: read it. Real safe harbor names CFAA + DMCA + state computer-crime statutes. Vague "we will not sue" language is worth less.
• VPN + dedicated machine. The BlackArch laptop already does the second; pin a paid VPN (Mullvad / IVPN) for the first. Some programs require US-only or geo-restricted source IPs; respect them or your reports get tossed.

#11. Concrete 12-Month Plan

A specific schedule that fits an OSCP-track schedule in parallel (see pentest.md for the cert side).

#Months 1-3 — Foundation

• Finish PortSwigger Web Security Academy start to finish (the apprentice + practitioner labs). It is free and it is the canonical curriculum.
• Set up the BlackArch laptop's recon stack (section 6) in ~/ops/. Write a small shell wrapper that takes a program slug and scaffolds ~/ops/<program>/{scope,recon,notes,reports}.
• Pick two public programs on Bugcrowd / HackerOne with broad scope and high signal-to-noise, usually big SaaS vendors with subdomain wildcards. Submit ten reports. Expect 6-8 dupes and 1-2 paid lows. This is normal.
• Read The Web Application Hacker's Handbook 2e chapters 4, 7, 9, 12, 13 in parallel with the labs.

#Months 4-6 — Specialize

You have two strong lanes given your background.

1. AI/LLM applied AppSec OpenAI (Bugcrowd), Anthropic (H1), Microsoft Copilot, Google AI VRP. Read OWASP LLM Top 10 + ATLAS, then hunt the AppSec surface (auth, IDOR, SSRF in tools, RAG leakage) on these targets. This compounds with your math/Python.
2. Smart-contract auditing on Immunefi + Sherlock work through Cyfrin Updraft and Secureum bootcamp materials. Do one Code4rena or Sherlock contest a month even if you score zero, the post-mortem reading is the real curriculum. This is the higher-payout lane.

Pick one. Doing both halfway is worse than doing one fully.

#Months 7-9 — Depth + relationships

• Target one program deeply. Build a recon-and-monitoring pipeline for it: weekly subdomain diff, JS-file diff, new-route diff. The 80/20 of finding criticals is being on the new attack surface first.
• Publish two writeups on krisyotam.com (cross-link from this doc). One technical, one methodology.
• Get on the program's Discord / Slack if they have one. Talk to triagers, humanise yourself. This becomes invite paths.
• Submit to the DEF CON AI Village CFP or a BSides talk. Even a rejected CFP gets read.

#Months 10-12 — Scale or invite

• If AI/LLM lane: aim for an Anthropic or Microsoft Copilot invite-only / private scope. Anthropic in particular has invited steady reporters into expanded scope.
• If smart-contracts lane: aim for a top-3 finish in one Sherlock or Code4rena contest, or land a Cantina solo audit invite.
• Either way: apply to a Live Hacking Event. HackerOne LHE applications happen rolling.
• Honest 12-month income target for a half-time hunter on this plan: $15k-$50k all-in. Hitting top-1% in year one is a fantasy; year three is realistic.

#Quick cross-links

• Cert path, OSCP/OSWE/OSEP, Synack vetting, salary data, formal pentest career: ./pentest.md
• Open source as income (sponsors, grants, bounties on OSS code you wrote): ./open-source.md
• Teaching / courses / writing: ./teaching.md, ./writing.md

#References

• HackerOne 7th Annual Hacker-Powered Security Report: hackerone.com/resources/reporting
• Bugcrowd Inside the Mind of a Hacker: bugcrowd.com/resources/reports
• Immunefi research + leaderboard: immunefi.com/research, immunefi.com/leaderboard
• OWASP LLM Top 10: genai.owasp.org/llm-top-10
• MITRE ATLAS: atlas.mitre.org
• DOJ CFAA policy update 2022: justice.gov/opa/pr
• DMCA section 1201 exemptions: copyright.gov/1201
• CISA Coordinated Vulnerability Disclosure: cisa.gov/coordinated-vulnerability-disclosure-process
• PortSwigger Web Security Academy: portswigger.net/web-security
• Cyfrin Updraft (smart contract security): updraft.cyfrin.io