The short answer
picoCTF is a free capture-the-flag hacking competition run by Carnegie Mellon University. Students are given deliberately broken software, encrypted messages, suspicious network captures and vulnerable web apps, and asked to break each one until a string of the form picoCTF{...} falls out. Find the string, paste it in, score the points. It ran for the first time in the spring of 2013, it has run in most years since, and since 2020 every challenge has stayed playable in a free archive after the season ends.
It was built by CMU's competitive hacking team, the Plaid Parliament of Pwning, together with a five-person game studio staffed by graduate students from CMU's Entertainment Technology Center. The reason it exists is written into the first sentence of the paper its founders published about it: "The shortage of computer security experts is a critical problem." picoCTF is a recruiting funnel for a profession, dressed up as a game about a broken robot.
That claim needs receipts, and there are two papers' worth. If you want to skip the history and start solving, the picoCTF beginner's guide is the ten-minute on-ramp and how to start playing CTFs is the wider version. Everyone else, stay. The story of how this competition got tuned is genuinely the most useful thing about it.
Who actually built it, and in what order
Three names are on the original paper: Peter Chapman, Jonathan Burket and David Brumley, all of Carnegie Mellon University. Brumley is the recognisable one. He was then the Gerard G. Elia Career Development Professor in CMU's Department of Electrical and Computer Engineering, and the faculty presence behind the Plaid Parliament of Pwning, the CMU team that has won the DEF CON capture-the-flag finals nine times as of August 2025, more than any team in that contest's history. Chapman and Burket were the graduate students who did the building.
The challenges came from PPP. That matters more than it sounds. picoCTF was not assembled by curriculum designers who had read about hacking; it was assembled by the people who keep winning the hardest hacking competition in the world, then filed down until a sixth grader could get through the first level. When CMU announced it in March 2013, Brumley described the premise like this:
"The story of the Internet competition begins when a robot from outerspace crash lands in your backyard, it's up to the game competitors to use their hacking skills to fix the robot and uncover its secrets."
The robot was not a metaphor. A second CMU team, five graduate developers from the Entertainment Technology Center working under the name Team Osiris, built an actual HTML5 adventure game around the challenges. It was called Toaster Wars. Four levels: you find the crashed robot in your backyard, you repair it, you recover its lost ship from a nearby spaceport, and you take it to space to enter an intergalactic hacking contest. Cut scenes, sound effects, the lot. Advanced players could switch to a plain text problem list, and roughly half of them did.
The first competition ran ten days in the spring of 2013. Registration required a teacher and a school, teams were five students or fewer, and students in grades 6 to 12 in the United States were eligible for prizes. Anyone else could play, they just did not appear on the scoreboard. Eleven organisations sponsored it, mostly to fund the prize pool, which ran from $1,000 to $8,000 in cash plus Amazon Web Services credit and a stack of Wiley books.
picoCTF 2013, by the numbers
| Measure | Value | What it tells you |
|---|---|---|
| Teams | 1,938 | From 955 separate schools, so most schools sent one or two teams |
| Challenges | 57 | Forensics 16, web and scripting 13, binary exploitation 11, reverse engineering 9, cryptography 8 |
| Answers submitted | 172,482 | Most of them wrong, which is how the organisers estimated time spent |
| Solved per team | 15 | Median 12. Three teams cleared all 57 |
| Hours played | 11 to 12 | Per student, over ten days, almost entirely outside class |
| Teams solving ROP1 | 96 | A return-to-libc attack, normally taught at graduate level, solved by high schoolers |
Why a university built a hacking game for teenagers
The paper states three goals, and they have not changed in thirteen years: encourage students to pursue computer science degrees, introduce security topics at a younger age, and give teachers material they can actually put in a lesson plan. Underneath all three sits a workforce number. A 2010 presidential commission had described a nationwide shortage of tens of thousands of security experts, and the founders' framing of the pipeline problem was blunt: fewer than 10% of US high schools offered the computer science Advanced Placement exam, and only 15,000 students sat it in 2011, against more than 200,000 who took AP Calculus AB alone.
CyLab, CMU's security and privacy institute, still frames it the same way. Their July 2025 write-up puts more than 700,000 unfilled cybersecurity jobs in the United States against a 230% rise in attacks since April 2020. Megan Kearns, the picoCTF program director, has been explicit that the gap starts before college: the United States has no standardised cybersecurity education for schools to teach in the first place, so something has to fill it.
The design choice that follows from that goal is the one people underrate. picoCTF is almost entirely offensive. The other high school competitions of the era, most visibly CyberPatriot, were defensive: harden this machine, pass this checklist. Brumley said in 2013 that defensive competitions "end up with competitors merely running through checklists," and that CMU's was "heavily focused on exploration and improvisation with elements of play."
A checklist teaches you what someone already knows is wrong. Breaking something teaches you to find what nobody has noticed yet. picoCTF bet the entire program on the second one.
I think that bet is the reason picoCTF works. A defensive exercise can only ask you to apply a rule someone else already wrote down. An offensive one makes you go looking for the case nobody wrote down. Both skills matter in a job, but only one of them is fun at fourteen, and fourteen is the entire point of this program.
The founders were also careful about the ethics of teaching attack first: the paper devotes a section to it, the competition shipped an introductory video on the subject, and the story was deliberately written so that every exploit helps a broken robot get home. That framing is doing real work and it is worth noticing, because the alternative reading of "we taught thousands of teenagers to write buffer overflows" is one a university has to answer for.
What 1,938 teams taught the people who built it
Here is where picoCTF stops being a nice story about a competition and becomes something more unusual. The organisers logged every submission, surveyed 394 students and 21 instructors afterwards, ran it past an institutional review board, and then published the parts that made them look bad.
The finding that reorganised the whole competition: students did not hate the hard challenges. They hated the challenges that required an unfamiliar tool.
Nearly half the participants asked for fewer binary exploitation challenges. Forensics and reverse engineering were unpopular too. Three of the most disliked individual problems were one that required editing a filesystem image in a hex editor and mounting it, one that required editing a cookie value, and one best solved with a Java decompiler. And then there is the detail I cannot get out of my head. One of the earliest, easiest problems in the whole competition was called Grep Is Your Friend. It asked students to find a string inside a set of files. It was the sixth most disliked problem in the competition.
# The entire intended solution to the sixth most hated challenge of 2013grep -r 'picoCTF{' files/
Meanwhile the three favourite challenges were an eval injection in Python, a set of equations to satisfy in another Python program, and an audio message in a thin disguise of Morse code. All three are moderately difficult. None of them requires a tool the student had never opened.
grep, because nobody had ever shown them a terminal.The authors resisted the obvious wrong conclusion, and I want to give them credit for it in their own words: they were "not suggesting that the distribution of challenges should be designed to match student preferences." The fix was not to remove the binary exploitation. The fix was to teach the tools first. Their stated plan for future competitions was to "separately introduce useful resources such as the command line, debuggers, and web developer tools."
Read a modern picoCTF season with that sentence in mind and the category list stops looking arbitrary. The 2013 competition had five categories and no General Skills. It is now a fixture, and in 2026 it was the largest category of the season at 17 challenges. Its opening challenges are named after the tools they exist to teach: Obedient Cat for cat, First Grep for grep, Nice netcat... for nc, strings it for strings, Wave a Flag for --help, and Python Wrangling for running a script someone else wrote. There is even a GDB Test Drive whose entire job is to make the debugger stop being scary.
The year picoCTF nearly scared everyone off
The tools problem was the first correction. The second one is better documented and, if you are learning from picoCTF right now, more useful.
After picoCTF 2017, the post-competition survey filled up with a specific complaint: the problems got too hard too quickly, and students lost confidence and gave up. Not "too hard." Too hard too quickly. The gradient was wrong.
Four people at CMU took that seriously enough to rebuild the 2018 competition around it and then write up the result. Kentrell Owens, then an electrical and computer engineering graduate student, led the paper with Alexander Fulton, CyLab software engineer Luke Jones, and Marty Carlisle, the education lead for picoCTF 2018 as a teaching professor at CMU's Information Networking Institute, and by the time the paper was published, at Texas A&M. They called it pico-Boo! How to avoid scaring students away in a CTF competition, and it won Best Student Paper at the Colloquium for Information Systems Security Education in 2019.
The intervention was unglamorous. Give every problem an explicit learning objective. Make each successive problem only slightly harder than the one before it. Add a classrooms feature so teachers can track their own students. Carlisle described the target as hitting "the sweet spot where students didn't get discouraged and give up, but also didn't get bored and quit because the problems were too easy."
What re-ordering the challenges did
| Season | Students completing 15+ problems | What changed |
|---|---|---|
| picoCTF 2017 | 25% | Difficulty jumped early; survey respondents reported losing confidence and quitting |
| picoCTF 2018 | 50% | Per-problem learning objectives, a graded ramp, and a classrooms feature for teachers |
By the following season the ramp was so deliberate that the people building it talked about it as a property of the product. Daniel Tunitis, the problem development lead for picoCTF 2019 and one of twelve Information Networking Institute students who wrote that year's challenges, put it plainly: "PicoCTF has a nice ramp of difficulty and gets progressively harder. Sometimes you just have to keep trying different things to figure out how a problem works."
I would push that further than he did. The ramp is not a nice property of picoCTF. It is the product. Any competent CTF team can write you 57 challenges. Almost nobody can tell you which order to attempt them in, because almost nobody has the drop-off data.
What picoCTF is in 2026
Thirteen years in, the thing has grown well past a ten-day event. Here is the arc, compressed.
2013
One competition, one game, one country
1,938 US teams play Toaster Wars over ten days. The platform is released as open source and adapted into the curricula of 40 high schools; six other capture-the-flag competitions run on it within a year.
2014 to 2016
A second run, then a two-year gap
The second edition runs from 27 October to 7 November 2014 with more than $30,000 in prizes and a new game team, Daedalus, in place of Osiris. Then nothing in 2015 or 2016. When it returns in March 2017, CMU is still calling it the third annual competition. What the group publishes during the gap is a paper on generating challenges automatically.
2017 to 2018
The difficulty curve gets engineered
The drop-off problem is identified and the ramp is rebuilt. picoCTF 2018 draws 27,221 participants from 154 countries, 13,596 of them eligible for prizes and 3,313 of them high school seniors. The sponsors thanked in that write-up are Boeing, Amazon Web Services, Eaton, Siemens, Ryerson University and CMU's own Information Networking Institute.
2019
Scale, and a year-round platform
More than 39,000 register for the September and October season, developed by twelve INI graduate students under faculty advisor Hanan Hibshi. picoCTF stops being an event you can miss and starts being a place you can visit.
2020
picoGym opens
Players had asked to keep practising after the scoreboard froze. picoGym makes the whole back catalogue permanently playable, which is why an archive site like this one can exist at all.
2024 to 2025
A platform with a competition attached
Around 4,400 US students compete over 47 challenges in 2024. By July 2025 CyLab counts more than 18,000 competitors that season and over 800,000 active users worldwide, alongside guided learning material and a monthly lecture series.
2026
Thirteen years on, and a new front door
picoCTF 2026 runs ten days from 9 March 2026, with more challenges than any prior year and a prize structure rewritten to spread awards across more teams rather than concentrating them at the top. CyLab puts the cumulative reach at close to a million learners.
The money question comes up a lot, so: it is genuinely free, and it stays free because CyLab funds it from industry sponsorship, foundations and government grants rather than from participants. Kearns puts it plainly: "In order to keep the platform free to all users, we rely on community funding, industry sponsors, foundations and government grants." The platform code is open source under the picoCTF GitHub organisation, though the repository that ran the 2019 competition is now archived and the current tooling lives in the challenge-manager projects beside it.
One more thing worth knowing before you register. Anyone in the world can play. Prize eligibility has always been narrower and division-based, historically centred on middle and high school students in the United States and, from at least 2018, Canada. Read the rules for the season you are entering rather than assuming. Playing is open. Winning money is not always.
What is actually on the exam
Six categories carry almost everything, with a seventh added in 2026. The archive on this site currently holds 513 solved challenges across twelve picoCTF events, which is enough to say what each category really asks for rather than what its name suggests.
| Category | What it really tests | Start here | Then read |
|---|---|---|---|
| General Skills | Terminal fluency, which the 2013 survey found was the real barrier | First Grep | General Skills roadmap |
| Cryptography | Recognising a scheme from its output, then attacking how it was used | Mod 26 | Cryptography roadmap |
| Forensics | Believing the bytes over the file extension | Glory of the Garden | Forensics roadmap |
| Web Exploitation | Trusting nothing the browser sends, because the server did | Insp3ct0r | Web Exploitation roadmap |
| Reverse Engineering | Reading code that was never meant to be read again | Transformation | Reverse Engineering roadmap |
| Binary Exploitation | Turning a memory bug into control of a running program | buffer overflow 0 | Binary Exploitation roadmap |
Compare that against the 2013 split and the drift is obvious. Forensics was the biggest category in the first competition at 16 of 57 challenges; across the full archive here it sits behind Reverse Engineering. Binary exploitation, the category half the 2013 field asked to see less of, is still there and still the smallest of the six. Nobody removed the unpopular thing. They put a ladder in front of it.
If you would rather work a category end to end than pick challenges at random, the learning paths on this site are ordered attempts at exactly that, and the event index groups every walkthrough by season if you are working a specific year.
The one thing the archive lost
Now the complaint, because a piece that only admires its subject is a brochure.
picoGym is the best thing picoCTF ever shipped and it quietly threw away the competition's most expensive asset. During a live season you meet the challenges in the order four researchers tuned for you, with a difficulty curve that doubled completion rates when they fixed it. In the archive you meet them as a catalogue you filter by category, event and difficulty, which is a different thing entirely: a filter answers "what is there?" and a ramp answers "what next?" The challenges survived. The sequence did not.
The archive kept every challenge and lost the one thing a paper was written about.
That is not a scandal, it is a product tradeoff, and permanent free access is worth more than a preserved ramp. But it does explain something that puzzled me for a long time: why so many people bounce off picoGym after loving picoCTF, and why third-party roadmaps and paths keep getting written for a platform that already has 800,000 users. They are all reconstructions of the same missing artifact.
So rebuild it deliberately. Pick one category. Do the tool challenges first even though they look beneath you, because the 2013 survey says that is where you will actually stall. Do not open a season index until you have finished the easy tier of one category, which is the mistake the ramp exists to prevent. And when you hit a wall, ask the pico-Boo question before the technical one: is this hard, or is this just unfamiliar?
picoCTF is thirteen years of a university writing down which of its students quit and why, then moving the furniture until fewer of them did. Play it in the order it learned.
Sources and further reading
Challenge and category counts come from the 513 picoCTF walkthroughs published on this site. Everything else is from CMU's own publications and press.
- PicoCTF: A Game-Based Computer Security Competition for High School Students, by Peter Chapman, Jonathan Burket and David Brumley, USENIX 3GSE '14. The source for every 2013 figure here, including the survey findings on tools and the challenge category split.
- pico-Boo! How to avoid scaring students away in a CTF competition, by Kentrell Owens, Alexander Fulton, Luke Jones and Martin Carlisle, Journal of The Colloquium for Information Systems Security Education. The source for the ramp methodology, the classrooms feature, and the 2018 participation and sponsor figures.
- How to teach cybersecurity without scaring students away, CMU CyLab, June 2019. CMU's own summary of that paper, and the source for the 25% to 50% change in students completing fifteen or more problems between the 2017 and 2018 seasons.
- Automatic Problem Generation for Capture-the-Flag Competitions, by Jonathan Burket, Peter Chapman, Tim Becker, Christopher Ganas and David Brumley, USENIX 3GSE '15. The origin of per-instance challenges and per-instance flags, and the source for the flag-sharing figures.
- CMU's 2013 competition announcement for the Plaid Parliament of Pwning and Team Osiris credits and the Brumley quotes.
- Team Osiris's own project page for Toaster Wars at CMU's Entertainment Technology Center, for the five-person game team and what it built, and CMU on the second annual competition (October 2014) for the 2014 dates, the prize total, and Team Daedalus.
- The Information Networking Institute's picoCTF 2019 report for the 2019 registration count, the student development team, and Hanan Hibshi's and Daniel Tunitis's comments on the difficulty ramp.
- CMU on making picoCTF more accessible (June 2024) and CyLab on the talent gap (July 2025) for the modern participation figures and Megan Kearns's comments on funding and standardisation.
- picoCTF returns in 2026 for the 2026 dates, the reworked prize structure, and the cumulative reach figure.
- picoCTF and CyLab Security Academy for the current platform, with the practice challenges themselves at learn.cylabacademy.org, plus the picoCTF GitHub organisation for the open-source platform code.