How to Become a Penetration Tester, Measured on 108 Open Postings

The route into penetration testing as employers describe it, counted from 108 offensive security postings read in full, with the skills, certifications, proof and pay each step calls for.

A penetration tester is paid to break into an organization's systems with its written permission, show how far a real attacker could get, and hand over a report that says what to fix first. The honest route in is slower than most course adverts suggest: a technical job beside security comes first (systems administration, software development, a security operations seat), then years of practice against deliberately vulnerable targets, a public trail of findings and write-ups, and usually one hands-on certification, most often the OSCP. Open postings confirm that shape. Of 108 offensive security postings this board read word for word on 11 November 2026, only 9 were internships or junior seats, the middle request among those that stated a number was five years of experience, and 45 named the OSCP, although not one of them made it the only way through the door.

That sample is the backbone of this guide. It holds the 87 genuine penetration testing and red team roles among the 93 the board files under that specialty that day (six were misfiled: five French postings where exploitation means operating a system, and one threat intelligence manager), plus 21 roles from other specialties whose work is attacking systems: vulnerability researchers, product security assessors, threat emulation consultants and AI red teamers. Together they come from 55 employers. Counts below are out of those 108 unless a sentence names another base, and all carry their date, because the board re-reads every employer's careers system each hour. As you read this, 97 penetration testing and red team roles are open on CyberJobs.cc, 39 of them remote, and the ones that state a salary sit in a typical band of $167k/yr to $320k/yr.

The search terms for this career are muddled, so a word on names. Penetration tester, pentester and ethical hacker describe the same core job; red teamer, offensive security engineer, vulnerability researcher and AI red teamer are related seats with different goals, and section 2 separates them. Where a figure comes from outside the board, the sentence links the page it came from.

Contents

1. What a penetration tester does, from scoping call to report
2. The kinds of offensive work, and who hires for each
3. What 108 open postings ask of applicants
4. The skills ladder, in the order it compounds
5. Certifications: what postings name, and the order to sit them
6. Building proof before anyone pays you to attack
7. Doors in: the jobs that lead to a first offensive seat
8. Pay at each level, from the salaries postings state
9. Where the legal line runs
10. AI as a new target and a new tool
11. A two-year plan from a technical job to a testing offer
12. Where the career goes after the first seat
13. Penetration tester questions, answered from the postings

1. What a penetration tester does, from scoping call to report

Start with why anyone pays for this work at all. A security team can list the controls it has bought, the patches it has applied and the policies it has written, but none of that proves an attacker cannot get in. Absence of a path is very hard to demonstrate from the inside, because the people who built a system share its blind spots. A penetration test converts "we think we are covered" into evidence: here is the route an outsider took, here is the data it reached, here is the step that would have stopped it. The buyer is purchasing a decision document. Everything else in the job, the tools, the exploits, the late nights on a stubborn login form, exists to make that document true and specific.

That framing explains the shape of an engagement better than any list of hacking techniques. NIST's long-standing guide to security testing, published in September 2008 and still current on its site, describes the work as planning and conducting technical tests, then analyzing what they find and developing ways to mitigate it - NIST SP 800-115. Planning comes first for a reason.

Scoping. Before any packet is sent, the tester and the client agree what is being tested and why. Which addresses, applications, cloud accounts or buildings are in, and which are explicitly out? Is the goal a broad sweep of one web application for every flaw it holds, or to see whether an outsider can reach the payment system? Is production fair game, or only a staging copy? Scoping is also where price is set, because a consultancy sells time, and the hours estimated here decide how deep the test can go.

Rules of engagement. The written agreement that turns hacking into a lawful service. It names the person with authority to approve the test, the testing windows, the techniques that are forbidden (denial of service is commonly excluded, social engineering often needs separate sign-off), how sensitive data found along the way is handled, who to call at two in the morning if something breaks, and what to do if the tester finds signs that a real attacker is already inside. Section 9 explains why the authority behind that signature matters so much.

Reconnaissance and testing. The visible part of the job: mapping what is exposed, enumerating services, users and endpoints, and then testing by hand for what automated scanners miss. Scanners are good at known vulnerabilities in known software; they are poor at business logic, broken authorization between users, and the chain of three individually minor issues that together hand over an administrator account. For web work, the OWASP Web Security Testing Guide, whose current stable release is version 4.2, is the reference methodology many teams follow - OWASP. For adversary-style work, MITRE ATT&CK supplies the shared vocabulary: a knowledge base of attacker tactics and techniques drawn from what real intrusions have done - MITRE ATT&CK.

Exploitation and restraint. Proving impact is the point, but proving it safely is the craft. A good tester demonstrates that a flaw reaches customer records by reading one harmless field, not by downloading the table. The same instinct runs through the products of this field: Horizon3.ai advertises its automated tests as production-safe, because a test that takes down the system it is protecting has failed its client.

Cleanup, report and readout. Every account, file and implant created during the test is removed and listed. Then comes the report: an executive summary that says in business terms what the findings mean, followed by each finding with its evidence, its severity, exact steps to reproduce it and a fix an engineer can act on. The readout meeting walks the client's team through it, and a retest weeks later confirms the fixes hold. Sophos's two senior penetration testing analyst postings list the rhythm plainly: kick-off calls, notification calls for high and critical findings while testing is still running, and close-out calls to review findings and evidence.

The consulting week against the in-house quarter

The same skills produce two very different working lives.

A consultant at a services firm meets a new client every week or two. Monday might be a scoping call for a bank's mobile application, the rest of the week an internal network test for a hospital, and Friday a readout for last week's retail client. Variety is the attraction, and the learning curve is steep because every environment is different. The constraint is the clock: the test ends when the purchased hours end, and the report has to be written either way. Consultancies also measure the business side of the work. Praetorian's posting for an offensive security manager lists utilization, gross margin and client satisfaction among the results the manager owns, which tells you exactly how a services firm sees a tester's week.

An in-house offensive engineer tests one company's products and infrastructure, again and again, at increasing depth. The work is less varied but goes further: a tester at Okta is expected to triage external bug bounty submissions alongside internal vulnerability tickets and to build tooling that outlives a single assessment, and the posting asks for a track record of shipping offensive tools beyond ad-hoc scripts. In-house testers also see their findings fixed, which consultants often never do.

Neither is the better start in general. Consulting compresses years of exposure into months and is where many testers learn the trade; in-house work rewards depth in one stack and, on this board, pays more (section 8). The postings in this sample split 43 in-house, 31 consulting or services, 23 vendor research and product roles, and 11 government-contract research roles, so both paths are well represented.

2. The kinds of offensive work, and who hires for each

"Penetration tester" is the search term, but the postings describe several distinct jobs that share a toolkit. Knowing which one you are aiming at changes what you study, which certification is worth its cost, and which employers to watch.

Penetration testing proper is time-boxed and breadth-first: the tester tries to surface every exploitable flaw in an agreed target, usually with the defenders informed. It is the bread and butter of consultancies, and the work most job seekers picture.

Red teaming pursues one objective (read the CEO's mail, reach the source code of the payment service) the way a specific adversary would, staying hidden for weeks, so the real test is whether the defenders notice. It leans on command-and-control frameworks and evasion of endpoint detection, which is why those terms cluster in red team postings. A close cousin, purple teaming, puts attackers and defenders side by side to tune detections in real time; 22 of the 108 postings mentioned it, from Applied Intuition's purple team engineer to GuidePoint Security's cleared threat emulation seat in Chantilly, Virginia.

Vulnerability research is the deepest specialization. Instead of testing a client's deployment of known software, a researcher hunts for flaws nobody has reported yet, often by reverse engineering binaries or firmware, and writes the exploit that proves them. Two Six Technologies alone had nine such postings open, every one asking for a US security clearance or the ability to obtain one, and security vendors such as Qualys, Palo Alto Networks and Check Point hire researchers to turn new exploits into product detections. The board's threat intelligence and research guide covers the defensive half of that research world.

Hardware, embedded and operational technology testing showed up more than outsiders expect: 29 postings named firmware, embedded systems, devices, vehicles, spacecraft or industrial control systems. Apex, which builds spacecraft, wanted debug-interface experience such as JTAG and CAN bus; Dragos wanted a principal OT penetration tester; Nebius wanted research into GPU firmware and drivers.

AI red teaming is the newest branch and is covered in section 10.

Four employment models, four different jobs

The table below sorts the 108 postings by who the tester works for, because that decides daily life more than the title does.

Three things stand out. Consulting is the most remote-friendly model (19 of 31), which fits work that happens over VPN connections and video calls. Consultancies almost never print pay (5 of 31 postings), so their salaries are the least visible part of the market. And government-contract research is the opposite on both counts: never remote in this sample, nearly always paid in the open, and gated by a clearance.

3. What 108 open postings ask of applicants

This section is the measurement. The counts come from reading each posting's full text, in English, French, Czech and Chinese where employers wrote it so, and counting a requirement whenever the posting names it, whether as required or preferred, since most postings blur that distinction.

Seniority: the market hires people who have done this before

By the level written in the title, 48 postings (44%) were senior, staff, principal or comparably senior seats, 41 (38%) carried no level word, 10 (9%) were lead or manager roles, 6 (6%) were internships and 3 (3%) were junior or associate seats. The internships were five French stages at Thales and one red team internship at CrowdStrike for students two years into a degree. The three junior or associate seats were an associate red team analyst at Armadin in Mexico, an associate security researcher at Cobalt and a pentest product associate at Wiz.

Missing level words should not be read as an open door for beginners. Among the 29 such postings that stated years, the middle request was three.

Years of experience: five in the middle, two at the floor

Seventy-seven postings stated a number of years. The middle was five, and 44 of the 77 asked for five or more.

The 12 postings at two years or fewer are the realistic first targets, and they are worth knowing by name. Praetorian's offensive security engineer posting asked for a degree in a technical field backed by zero to two years of experience or relevant internships, and wanted notable results in capture-the-flag competitions. Bitdefender's two penetration testing consultant postings, in Germany and Romania, asked for a minimum of two years in cybersecurity. Armadin's associate analyst seat accepted one year gained through internships, classes or projects. Wiz's associate seat asked for two years spent in application security or testing work. The years can come from adjacent work, which is the point of section 7.

At the top, the lead and manager seats asked for a middle of seven and a half years, and Armadin's managing principal red team operator asked for twelve.

Certifications: named often, required alone never

Forty-eight postings (44%) named at least one offensive certification. The OSCP led with 45, far ahead of anything else; section 5 sets out the full ranking and what it means. The striking detail is how the OSCP was named. In none of the 45 postings was it the single credential a candidate had to hold. It sat in lists of acceptable alternatives ("OSCP, OSWE, OSEP, GXPN, GPEN, or CRTO"), under preferred or bonus headings, or beside "or equivalent experience". Bishop Fox's posting said such certifications "can be helpful, but are not a necessity".

Skills and tools

Programming is the common language. Python appeared in 64 postings (59%), and 82 (76%) named at least one programming or scripting language. Go followed with 31, C or C++ with 23, PowerShell with 21 and Bash with 20. Offensive work has moved from running tools to writing them: Stripe wanted "strong programming skills in Python, Go, or similar languages" with demonstrated experience building tools, and Hadrian wanted offensive tooling built in Python, Go or C and C++.

Web and API testing is the largest single target. Web applications appeared in 48 postings (44%) and APIs in 38 (35%). Burp Suite, the intercepting proxy most web testing runs through, was the most-named tool at 34 postings.

Cloud has caught up with the network. AWS, Azure or Google Cloud was named in 46 postings, and Kubernetes in 29. By contrast, Active Directory and its usual attack tooling (Kerberos attacks, BloodHound, Entra ID) appeared in only 14. That gap says more about who posts on this board than about the field: the board reads software companies, cloud providers and AI model makers heavily, and their testers spend their days on applications and cloud accounts. Internal network testing against Active Directory remains the core of consulting work and of the OSCP exam itself, so do not read 13% as permission to skip it.

Deeper specialisms are common, not exotic. Exploit development appeared in 35 postings, reverse engineering in 33, disassemblers such as IDA, Ghidra or Binary Ninja in 15, and fuzzing in 17. Some form of command-and-control framework (Cobalt Strike, Sliver, Mythic or a generic C2) was named in 24, Metasploit in 23, and evasion of defensive tooling such as endpoint detection in 8. Mobile testing appeared in 22.

Writing, travel, clearance and degrees

Report writing is nearly universal. Seventy-eight postings (72%) asked for written reports, documented findings or strong written communication. ESET's Czech posting went further: its selection process may include a take-home challenge that tests both finding bugs and writing them up.

Travel is real but modest. Seventeen postings set a travel requirement, almost all at consultancies and security vendors' services arms, typically "occasional" or up to 10 to 25%. The exception was Dragos, whose principal OT penetration tester had to be willing to travel 30 to 40% of the time, at home and abroad.

Clearances are concentrated. Twelve postings asked for a US security clearance or eligibility for one, nine of them at Two Six Technologies; four more Thales postings in France and Canada said the hire would go through national security vetting. If you hold a clearance, the cleared roles page is the shortest list of employers to watch.

Remote work is common. Forty-two of the 108 (39%) were listed as remote. The board's remote offensive security list counted 35 of the specialty's 93 roles as remote on 11 November, with 17 of them open only to applicants in the United States and 4 open to any country.

Degrees are mentioned, and often waived. Forty-seven postings named a degree, and 27 of those offered equivalent experience as an alternative. SpaceX's red team posting accepted two years of professional experience in place of a degree; Palo Alto Networks' internal testing role traded a bachelor's plus five years against a master's plus three.

Public proof is asked for more often than any certification except the OSCP. Forty-nine postings (45%) asked for some public trace of offensive work: published research, conference talks, CVE credits, bug bounty results, capture-the-flag placings or open-source tools. Section 6 is about building exactly that.

4. The skills ladder, in the order it compounds

The postings list skills as if they were independent. They are not. Each layer of offensive skill is built on the one beneath it, and learning them out of order is the most common reason people stall. Here is the ladder from the bottom, with the reason each rung comes where it does.

Rung 1: networks and protocols

Every attack is traffic. Before exploits mean anything, you need to read what moves between machines: how DNS resolution works and fails, what a TCP handshake looks like, where TLS sits, what an HTTP request and response actually carry, and which services listen on which ports. Without this, scanner output is noise and a firewall rule is a mystery. A good test of the rung: capture your own browsing traffic and explain every packet in a page load.

Rung 2: operating systems, from the inside

Linux appeared in 46 postings and Windows in 26, and both are needed. The goal is not to use them but to understand how they decide who may do what: users and groups, file permissions, services and how they start, scheduled jobs, where credentials are stored, and how a process inherits rights. Privilege escalation, named in 13 postings, is simply the craft of finding where those decisions were made carelessly. Testers who ran servers for a living climb this rung fastest, which is one reason administrators make strong testers.

Rung 3: a programming language you can build with

Python first, because 64 postings name it and because it lets you automate every repetitive step that follows. The bar is higher than "can read a script". Postings ask candidates to write custom tools, adapt public exploits, parse odd protocols and build harnesses. After Python, pick a second language by direction: Go for tooling and cloud work, C and C++ for exploit development and research, PowerShell or C# for Windows and red team work.

Rung 4: web applications and APIs

The largest target in the postings, and the rung where most people get their first real findings. Learn how sessions and tokens work, then the classic flaw families: injection, cross-site scripting, server-side request forgery, insecure deserialization, and above all broken access control, where one user can read another's data by changing an identifier. That last class is why testers who think like developers find so much: the bug is in logic, not in a library. Spend serious time in an intercepting proxy. The application security guide describes the defensive side you will be reporting to.

Rung 5: identity and Active Directory

Corporate networks still run on Active Directory, and internal tests still end at a domain administrator account. Learn how Kerberos and NTLM authenticate, why service accounts with weak passwords are a gift to attackers, how permissions on directory objects create paths no administrator intended, and how graph tools map those paths. Then learn the cloud version of the same idea: identity providers, single sign-on and the hybrid links between on-premises directories and cloud tenants. OffSec's exam for the OSCP gives a whole AD set of three machines 40 of its 100 points, which shows how central the testing world still considers this rung - OffSec OSCP+ exam guide.

Rung 6: cloud and containers

With 46 postings naming a cloud provider and 29 naming Kubernetes, cloud testing is no longer a niche. The attacks are mostly identity attacks wearing new clothes: over-broad roles, leaked access keys, metadata services that hand out credentials, trust relationships between accounts, and containers that can escape to their host. It sits above rung 5 because cloud identity is easier to understand once you have seen how directory permissions go wrong.

Rung 7: one specialization, chosen deliberately

Only at this point does specialization pay off, and the postings show where it pays: exploit development and reverse engineering for research roles, command-and-control and detection evasion for red teams, mobile for application testers, hardware and firmware for device and aerospace work, AI systems for the newest seats. A generalist with rungs one to six can apply to most of the consulting and in-house postings in the sample; a specialist with all seven is who the research labs and red teams were hiring.

The rung that runs alongside all of them: writing

Seventy-eight postings asked for it, and it is the skill most self-taught testers neglect. A finding nobody can reproduce is worth nothing to the client. Practise writing every lab exercise up as if a stranger had to fix it: what you found, why it matters in plain business terms, the exact steps, the evidence, and the fix.

5. Certifications: what postings name, and the order to sit them

Certifications are where aspiring testers spend the most money, so it is worth seeing exactly what employers write. The chart counts the 108 postings that named each certification as required, preferred or acceptable.

The OSCP is the benchmark, and here is why

The OSCP was named in 45 of the 108 postings, nearly three times as often as any other. Across the whole board it appeared in 119 of 2,150 open security postings that day (5.5%), most often in pentest and red team postings (36), then security engineering (24) and application security (19), according to the OSCP page.

The reason is the exam's design. Its current form, the OSCP+, gives candidates 23 hours and 45 minutes in a private network of vulnerable machines, then another 24 hours to upload a written report; the network holds three standalone machines worth 60 points and an Active Directory set of three machines worth 40, and every attempt is proctored - OffSec OSCP+ exam guide. The same guide says the credential is awarded without prerequisites to anyone who passes. In other words, the certificate is a performance test that mirrors the job: work alone under time pressure against unfamiliar systems, then document it well enough to be graded. For a hiring manager screening hundreds of applicants, that is a cheap and reasonably honest signal, which is why postings reach for it.

Two caveats follow from the counts. First, it was never the only acceptable route in this sample, so a strong portfolio can stand in for it, and some employers say so. Second, the board-wide salary attached to OSCP postings (a middle of $196,000 on the OSCP page) reflects which jobs name it, not what the certificate adds to anyone's pay.

The advanced tier

The next band of names is the advanced OffSec family, which together appeared in 35 postings: OSWE for web exploitation (18), OSCE or OSCE3 (17), OSEP for evasion and red team tradecraft (14) and OSED for exploit development (5). Across the board, 65 open postings named one of them, as the advanced OffSec certifications page shows. GIAC's offensive certifications came next, in 24 postings: GXPN for exploit work (17), GPEN for network testing (15) and GWAPT for web applications (10). The CRTO, a red team operations certification, appeared in 13, nine of them red team or consulting postings; the board counted 14 across all its open roles on the CRTO page.

These are second and third certifications. They are worth sitting once the direction is chosen: OSWE or GWAPT for an application tester, OSEP or CRTO for a red team operator, OSED or GXPN for someone heading toward research.

The entry tier is barely named

Certifications marketed to beginners hardly appear. PNPT and CRTP were named in 5 postings each, the eJPT family in 3, the CEH in 4, CompTIA PenTest+ and Security+ in 2 each. Where they did appear, it was often in long lists of acceptable options, as in Delinea's posting naming fifteen certifications any one of which would do, or at the associate level, as in Cobalt's associate researcher posting, which listed Security+ and several practical entry certifications alongside "working toward an OSCP".

The CEH deserves a direct word because so many people ask about it. Across the whole board it was named in 59 postings against 119 for the OSCP, 17 postings named both, and the CEH was asked for most in security engineering rather than in testing roles, as the OSCP and CEH comparison records. Where it does matter is US defense contracting: GuidePoint Security's threat emulation seat cited the Department of Defense baseline, for which the CEH and Security+ were its examples.

A sensible order

Read this as a default, not a rule. If your first job is in IT or a SOC and the employer is a defense contractor, Security+ satisfies the baseline those postings cite and is often paid for. For the testing career itself, the OSCP is the first certification the postings reward, sat once rungs one to five are solid rather than as a way to learn them. Afterwards, one advanced certification in your chosen specialization. And throughout, the portfolio in section 6, which 49 postings asked for, matters as much as any of them. The certifications page shows today's counts for every certification across all ten specialties.

6. Building proof before anyone pays you to attack

The central problem of breaking in is circular: employers want experience attacking systems, and attacking systems without permission is a crime. The postings show how the field resolves it. Forty-nine of the 108 asked for public evidence of offensive work, and that evidence can be built entirely legally, on your own time, before the first offer.

What the postings counted as proof

CVE credits were named in 25 postings, bug bounty results in 14 and capture-the-flag results in 10, alongside conference talks, published research, blog posts and open-source tools. The wording is revealing. Bishop Fox's posting for a managing consultant imagined that the candidate "may have accumulated a few disclosures, written blog posts, and given some industry talks". Anthropic's red team engineer posting asked for "a public body of work such as CVEs, blog posts, or disclosed bug bounty reports". Bitdefender's consultant postings looked for "involvement in security initiatives outside of work, such as public disclosure of vulnerabilities or relevant awards from bug bounty programs". One consultancy's application form asked for a link to the applicant's profile on a practice platform for deliberately vulnerable machines.

Five kinds of proof, in rising order of weight

1. A home lab you can describe in detail. Build a small Windows domain with a domain controller, a couple of member machines and a deliberately weak configuration, plus a few intentionally vulnerable web applications. Attack it, then switch sides and look at what your attacks left in the logs. The second half is what makes you useful to a purple team and impressive in an interview.

2. Write-ups of retired practice targets and competition challenges. Practice platforms publish retired machines that may legally be attacked and openly discussed. A write-up that explains why each step worked, what you tried that failed, and how a defender would have caught you shows thinking, not just a flag. Ten good write-ups beat a hundred unexplained completions.

3. Capture-the-flag placings. Competitions reward speed and breadth, and team events teach collaboration under pressure. Praetorian's posting named two collegiate competitions specifically, one defensive and one a penetration testing contest, which is a useful hint for students.

4. Real findings through bug bounty and disclosure programs. Here you test real systems, but only those whose owners have published a program inviting it, and only inside the scope that program defines. Since a US directive of 2 September 2020, every federal agency it covers has had to publish a vulnerability disclosure policy that commits the agency not to pursue legal action against good-faith research that follows it, and to treat such research as authorized - CISA BOD 20-01. Read each program's rules before you start, keep evidence, and report through the channel it names. A first accepted report, even a low-severity one, is the most persuasive line a career changer can put on a resume.

5. CVEs and published research. A CVE identifier is the public catalogue number a reported flaw receives once the vendor or a coordinator confirms it. Finding one usually means auditing open-source code or a product you can run yourself, reporting it privately, and waiting for the fix before writing about it. It takes longer than a bounty but carries more weight, because it shows you can find something nobody else had.

Make tools, not just findings

Several postings asked for tooling: Okta for "a track record of building and shipping offensive security tooling beyond ad-hoc scripts", Armadin for tools "others actually use". A small open-source tool that automates a tedious step in your own testing, documented and maintained, is proof of both programming and judgment about what is worth automating.

Turn it into an application

The cybersecurity resume guide shows how to present engagements, findings and severities on a tester's resume without exposing any client. For the interviews themselves, offensive loops often include a practical exercise and a methodology question; the interview questions guide works through typical attack-path questions, and the interview practice tool lets you rehearse a technical round with feedback on each answer.

7. Doors in: the jobs that lead to a first offensive seat

With 9 entry-level seats among 108 postings, most testers arrive from somewhere else. The board's offensive security career guide estimates two to three years before a first role, most often reached through IT or the SOC. The useful question is therefore not "how do I start in penetration testing" but "which neighbouring job teaches the rung I am missing, and leads most directly to a testing seat".

From running servers and networks. Administrators already hold rungs one and two, and often five: they know how a domain is configured because they configured one. The natural first target is infrastructure and internal network testing, where knowing how administrators cut corners is half the job. The move is strongest toward consultancies, whose internal tests run on exactly that knowledge.

From software development. Developers hold rungs three and four and read code fluently, which is what application testing and source code review need; 24 postings asked for code review skills. The common path runs through application security first. Bishop Fox's penetration tester posting is a good illustration of how employers think about it: four years of web application testing and five of application security, plus code review skill, with the OSCP only a bonus.

From security operations and detection. SOC analysts know what attacks look like from the defender's chair, which is exactly what red and purple teams need to understand in order to evade it, and what consultancies need when they explain findings to defenders. Purple team seats, such as the ones at Applied Intuition and GuidePoint Security in the sample, are a natural bridge. The security operations guide describes the seat this door starts from.

From vulnerability management. Running scanners, triaging their output and chasing fixes teaches which findings matter and why. It is a short step from proving a scanner wrong to proving a flaw exploitable, and several postings blended the two.

From cloud engineering. With 46 postings naming a cloud provider, an engineer who has built cloud accounts and Kubernetes clusters holds a skill many testers lack. Cloud testing seats at infrastructure companies were among the better paid in the sample.

From military or government service. A clearance is a door of its own. The nine Two Six Technologies postings and the defense-sector seats at Anduril Industries and Apex all asked for one or for eligibility, and a veteran who already holds it skips the long wait employers otherwise face.

From a classroom. Students have internships, and they are scarce and early. The CrowdStrike red team internship in the sample was for summer 2027, open in November 2026; the internships page lists every current internship and new-graduate security role.

Where juniors are actually trained

Junior penetration testing seats exist where someone senior can review the work. That points to consultancies with enough volume to pair a new hire with experienced testers, to the services arms of security vendors, and to large in-house teams. Junior testing seats are rare even on a board built to surface them: 4 of the 60 junior security roles open on 11 November were in penetration testing and red team work, and the list of starting-rung roles notes that such seats tend to want certifications or documented lab work. A small in-house team of two or three testers rarely hires a beginner, however much it needs help, because nobody has time to check a beginner's findings before they reach engineers.

If you have no technical job yet, the first step is the IT or SOC seat, not a testing seat, and the board's guide to getting into cybersecurity maps that first move in detail.

8. Pay at each level, from the salaries postings state

Pay figures for this specialty are thinner than for most, because the employers who hire the most testers, consultancies, rarely print a salary. In the sample, 47 postings from 27 employers stated a yearly range. Their middle low end was $160,000, their middle high end about $237,000, and the middle of their midpoints about $197,000.

By level

The level in the title mattered less than you might expect. Postings with no level word had a middle midpoint of about $206,500 across 18 ranges, and senior, staff and principal postings about $194,000 across 26. The no-level group was lifted by AI model makers that title their engineers without levels. No internship or junior posting in the sample stated a yearly salary, and the board's entry-level page had too few junior testing salaries to compute a band.

The board's own penetration tester salary page, which reads every role in the specialty, put the picture the same way on 11 November: across 41 distinct ranges from 25 employers, a typical $169,000 to $320,000 with the middle at $200,000; titles with no level word at a middle of $200,000, senior titles $201,000, and lead or manager titles $216,000. Remote roles stated a middle of $222,000; roles worked on-site or hybrid, $193,000. The highest-paying cybersecurity jobs ranking put penetration tester or red teamer sixth among eighteen jobs, with a middle of $213,000.

By employer

Who employs you matters more than your title. The chart shows the middle salary by kind of employer from the board's pay page.

The pattern follows from what each employer is buying. A company building AI models, or a large product company, pays a tester to protect a business worth far more than the salary, and competes with every other well-funded company for the same few people. A consultancy sells the tester's hours to clients, so salary is bounded by the billing rate minus margin, which is why consulting pay sits lower and why its postings so rarely print it. Defense research is paid on government contract rates, often lower in salary but stable and cleared.

The extremes in the sample make the spread concrete. The lowest stated range was $84,387 to $126,581 for a vulnerability researcher at Two Six Technologies in Ohio; the highest were $320,000 to $405,000 for Anthropic's red team engineer working on model safeguards and $220,000 to $405,000 for an offensive security engineer at Perplexity.

The economy-wide baseline

The US Bureau of Labor Statistics publishes no separate figure for penetration testers. Its closest occupation, information security analysts, held about 192,900 jobs in 2025 with a median annual wage of $129,180 in May 2025, and is projected to grow 21% between 2025 and 2035 - BLS Occupational Outlook Handbook. The gap between that median and the board's figures is not a contradiction. The government figure covers every employer and every level of analyst; the board reads mostly technology employers and, among testers, mostly experienced ones. A first offensive role at a regional consultancy will pay well below the board's middle, and that is a fair offer for that rung. The salary calculator sets a specific offer against what open postings state for the specialty and level.

9. Where the legal line runs

Only one thing separates a penetration test from a computer crime: authorization from someone entitled to give it. Every professional habit in this field, from the signed rules of engagement to the scope list taped beside a monitor, exists to keep that line bright. This section explains where the line sits in the two jurisdictions most readers will meet, and the case that showed how a signature can still fail. It is general information, not legal advice.

United States

The federal Computer Fraud and Abuse Act makes it an offence to intentionally access a computer "without authorization" or to exceed authorized access and thereby obtain information - 18 U.S.C. 1030, via Cornell Law School. In 2021 the Supreme Court narrowed the second phrase: a person exceeds authorized access when they reach parts of a computer, such as files, folders or databases, that are off-limits to them, not when they misuse access they legitimately have - Van Buren v. United States, decided 3 June 2021. For a tester, the practical reading is that the scope document defines which gates are open.

The Department of Justice added a charging policy on 19 May 2022. It defines good-faith security research as accessing a computer solely to test, investigate or correct a security flaw, done in a way designed to avoid harm, with the results used mainly to improve security, and says such research should not be charged under the act; the same activity used for other ends, such as extortion with discovered flaws, is not covered - Morrison Foerster summary. It is a prosecutorial policy, not a law, and it does not shield anyone from civil claims or from state law.

United Kingdom

The Computer Misuse Act 1990 makes it an offence to cause a computer to perform any function with intent to secure access to a program or data when that access is unauthorized and the person knows it, with up to two years in prison on indictment for the basic offence - legislation.gov.uk. Most other countries have equivalents, and a test that crosses borders, as cloud and remote work make common, may fall under more than one.

Third parties own part of every target

A client can authorize testing of its own systems, but much of what it runs belongs to someone else: a cloud provider, a software-as-a-service vendor, a hosting company. Each sets its own rules. AWS, for example, lets customers test their own infrastructure without prior approval for a listed set of services, but prohibits activities such as denial of service and DNS zone walking, requires red team exercises to go through its simulated events process, and requires prior approval for any testing that includes command-and-control - AWS. Checking those rules is part of scoping, not an afterthought.

The courthouse case

On 11 September 2019, two testers from the consultancy Coalfire were arrested inside the Dallas County Courthouse in Adel, Iowa, during a physical security test that Iowa's court system had contracted through its State Court Administration. The county sheriff argued that the court system's contract was invalid for a county building, a later review questioned whether the administration had authority to approve out-of-hours physical testing of courthouses, and the trespassing charges were eventually dropped - GovInfoSecurity, 3 February 2020. The two later received a $600,000 settlement after suing over the arrest - Reason, 17 February 2026.

The lesson is not that physical testing is reckless. It is that authorization must come from whoever actually controls the asset, and that a contract signed by the wrong party protects nobody at three in the morning. Ask, for every target in scope, who owns it and whether the person signing can speak for them.

Habits that keep a tester safe

Carry the authorization letter, with an emergency contact who will answer. Stop and call the moment something falls outside scope. Never test your own employer's systems, or a former employer's, without written sign-off from someone with authority, however obvious the flaw. Treat data you encounter as the client's, touch no more of it than proof requires, and record what you saw. On bug bounty programs, read the scope twice and stay inside it. These habits are also what interviewers probe when they ask how you would prove an injection flaw's impact without causing harm.

10. AI as a new target and a new tool

Artificial intelligence turned up in offensive postings in two distinct ways, and a newcomer should understand both.

AI systems as something to attack

Twenty-eight postings (26%) made testing AI systems part of the job: prompt injection, jailbreaks, abuse of the tools an agent can call, leakage of data through a model, and attacks on the infrastructure that serves models. Some are dedicated seats. Coinbase posted an offensive security engineer role focused on AI that includes red teaming its AI systems; OpenAI posted a novel abuse red teamer and a cyber red team role; Anthropic's red team engineer for safeguards wanted experience in model jailbreaking and in testing agentic workflows for prompt injection. Others fold AI into a general testing job: Forward Financing's senior penetration tester would test the company's core systems "along with the AI systems and agents our engineers are shipping".

The underlying skills are not new. An AI agent that can call tools is an application with an unusually persuadable input parser, and testing it draws on the same instincts as testing for injection and broken authorization: what can the attacker control, what can the system reach, and what happens at the boundary. A tester with solid web and API skills who learns how language models fail is well placed. The board's AI security roles page counted 231 roles whose work is securing AI systems, whose typical stated pay ran from $177,000 to $350,000, and a separate page follows the AI red teaming and security evaluation employers, where short engagements can give a tester first-hand experience of attacking language models.

AI as part of the tester's own kit

Forty-two postings (39%) expected the tester to use AI models or agents in their own work: to speed up reconnaissance, triage findings, write exploit code, or run testing at a scale no team could cover by hand. Liftoff wanted someone to lead adoption of AI-assisted offensive security tools; Palo Alto Networks' internal testing role would integrate AI-assisted testing frameworks into penetration tests. Several employers in the sample sell the automation itself: Horizon3.ai's platform runs autonomous penetration tests, Armadin describes building an "agentic attacker swarm", and depthfirst builds AI agents that find unknown vulnerabilities in large codebases. Each of them was hiring experienced offensive people to build and check those systems.

What that means for a beginner

Reason from what is being automated. The parts of a test that follow known patterns, such as enumerating services, matching software versions to known vulnerabilities and running standard checks, are exactly what automation absorbs first, and that part of the junior workload is shrinking. What remains human is judgment: deciding what is worth testing, chaining small weaknesses into a path that matters to this business, finding a flaw nobody has catalogued, knowing when to stop, and explaining risk to a person who has to fund the fix. Those are the skills in rungs four to seven and in the writing rung beside them. A newcomer who can direct automated tools and then verify what they report is more employable than one who competes with them at running scans.

11. A two-year plan from a technical job to a testing offer

This plan assumes you already work in a technical role (support, administration, development or a SOC) and can give offensive study about ten hours a week. If you do not yet have a technical job, make getting one the first goal and start this plan alongside it. Adjust the order to the rungs you already hold.

Months 1 to 4: networks, operating systems and Python. Close any gaps in rungs one to three. Set up the practice domain and vulnerable web targets from section 6, and log every command you had to look up twice. The checkpoint at month four: capture and explain a page load, escalate privileges on a deliberately weak Linux machine, and write a Python script that automates one dull task from your current job.

Months 5 to 9: web applications and your first write-ups. Work through the main vulnerability classes on deliberately vulnerable applications, using an intercepting proxy for everything. Publish your first five write-ups. Read one public penetration test report or methodology a week to absorb how professionals structure findings. If you are a developer, volunteer for security reviews in your own team.

Months 10 to 15: Active Directory, cloud and the OSCP. Build the Windows domain in your lab and work through the common attack paths, then repeat the exercise in a free-tier cloud account. Practise against retired machines under time limits, writing a full report for each as if a client would read it. Book the OSCP exam once full practice networks feel routine rather than as a way to force the learning.

Months 16 to 20: real findings. Choose one or two bug bounty or vulnerability disclosure programs with broad scope and spend steady time on them, reading their rules first. Aim for one accepted report, not dozens of duplicates. If you prefer depth, audit an open-source project for a CVE instead. Give a short talk at a local security meetup about something you learned.

Months 18 to 24: applications. Apply before you feel ready. Target the postings that ask for two to three years and accept adjacent experience: consultancies' associate and no-level testing seats, application security seats with testing duties, purple team roles, and internal moves in your own company. Use the board's offensive security roles list to see which employers are hiring right now, and rehearse the practical round of the interview.

Two years is not a promise. People who already administer Windows domains or ship web applications often move faster; people starting from a support desk often need longer. What the plan does guarantee is that the time goes into the things the 108 postings actually asked for, in an order where each step makes the next one easier.

12. Where the career goes after the first seat

The sample shows three directions, and they can be combined over a career.

The technical track. Most of the 48 senior, staff and principal postings were for hands-on testers and researchers, not managers. A principal consultant at Palo Alto Networks' Unit 42, a principal OT penetration tester at Dragos with eight years' experience, a principal security researcher at GitLab with ten: the most senior testers keep testing, lead the hardest engagements, set methodology and mentor others. The highest-paying jobs ranking places principal and staff security engineers third of eighteen, which is where many senior offensive engineers sit.

Management. Ten postings were lead or manager seats. DoorDash's engineering manager for its red team asked for seven years of offensive experience including three managing offensive practitioners, at a stated $193,800 to $285,000, remote. Praetorian's offensive security manager would lead a team of up to ten offensive engineers and own the commercial results of their work. Managing testers means scoping, staffing and quality control of reports, and less hands-on hacking.

Research and tooling. The deepest technical path leads into vulnerability research and exploit development (the Two Six Technologies, Qualys and Palo Alto Networks research postings), or into building offensive products. Horizon3.ai's senior and staff software engineers for its penetration testing product asked for six and eight years of software engineering, with security experience as a plus. AI red teaming is the newest destination, and section 10 shows it already pays at the top of the market.

Offensive experience also travels well outside offense. Testers become application security leads, security architects and heads of product security, because few people understand how systems fail as concretely as someone who has spent years making them fail. The application security guide describes one of the most common next homes.

13. Penetration tester questions, answered from the postings

From a first IT job, how long until a penetration testing offer?

Usually two to three years from a first technical job, and longer from a non-technical start. The board's offensive security guide puts the first role at two to three years, often by way of IT or the SOC, and the postings agree: the middle request among the 77 that stated years was five, and only 12 asked for two years or fewer.

Can you become a penetration tester with no experience?

Not directly, in most cases. Only 9 of 108 postings were internships or junior seats, and the junior ones still wanted a year or two of hands-on work, internships or lab projects. The workable route is a technical job first, as section 7 describes, with lab work, write-ups and bug bounty results built alongside it.

Do you need the OSCP to get hired?

It helps more than any other certification, but no posting in the sample required it alone. It was named in 45 of 108 postings, always as one acceptable option, a preferred item, or beside "or equivalent experience". A strong portfolio of findings and write-ups was asked for in 49 postings and can carry the same weight.

How much does a penetration tester make?

On this board's postings, which lean toward technology employers, the middle stated salary for the specialty was $200,000 a year on 11 November 2026, within a typical $169,000 to $320,000, and today's band is $167k/yr to $320k/yr. Consultancies, which hire many first-time testers, pay less and rarely publish ranges. Across all US employers, the closest government figure is the $129,180 median for information security analysts in May 2025.

Do penetration testers need a degree?

Often not. Forty-seven of the 108 postings named a degree, and 27 of those accepted equivalent experience instead. Defense and research employers are the most likely to insist on one; consultancies and product companies care more about demonstrated skill.

Is penetration testing a remote job?

Frequently. Forty-two of the 108 postings were remote, and consultancies were the most remote-friendly employers. Expect occasional travel at consultancies (17 postings set a requirement, usually under a quarter of the time) and on-site work for cleared research and hardware testing.

What is the difference between a penetration tester, an ethical hacker and a red teamer?

A penetration tester and an ethical hacker do the same job under different names: authorized, time-limited attempts to uncover every exploitable flaw in an agreed target. A red teamer pursues a single objective as a real adversary would, quietly, over a longer period, to test whether defenders detect and respond. Red team postings lean harder on command-and-control frameworks, evasion and Active Directory.

Is penetration testing legal?

Only with written authorization from someone entitled to give it, inside an agreed scope. Without it, the same actions are crimes under laws such as the US Computer Fraud and Abuse Act and the UK Computer Misuse Act. Section 9 covers where the line sits and the courthouse case that showed why the authority behind a signature matters.

Will AI replace penetration testers?

It is replacing parts of the work, not the job. Twenty-eight postings hired people to test AI systems and 42 expected testers to use AI tools; several employers selling automated testing were hiring experienced testers to build it. The routine, pattern-matching part of testing is automating fastest, which raises the value of judgment, chaining, novel findings and clear reporting.