Pwnagotchi Combines Artificial Intelligence with a Cute Pet to Capture the World's Wi-Fi Handshakes

Simone Margaritelli has released version 1.0.0 of Pwnagotchi, a Raspberry Pi-powered Tamagotchi-like virtual pet with a difference: it 'eats' Wi-Fi handshakes, using artificial intelligence to get smarter at feeding itself from nearby networks.

"This is the story of a summer project that started out of boredom and that evolved into something incredibly fun and unique," Margaritelli writes by way of introduction. "It is also the story of how that project went from being discussed on a porch by just two people, to having a community made of almost 700 awesome people (and counting!) that gathered, polished it and made today’s release possible."

That project: Pwnagotchi. Originally a simple pocket-size tool for carrying out Wi-Fi de-authentication attacks in the hope of capturing handshakes which could be used to break the security of wireless networks, the project took an abrupt shift when it gained a face. "I had no idea back then that just adding a simple, ASCII based face to something was the best way to get emotionally overly attached to that thing," Margaritelli explains. "I also wasn’t expecting another effect that showed up from the beginning: by giving it different 'moods,' and by having those moods depending on a real world environment, I created a Wi-Fi-based automata whose mood transitions were everything but trivial."

A personality followed the face, and a rudimentary artificial intelligence followed the personality. "There are two relatively simple neural networks that at each epoch are trying, in a way competitively, to estimate how the current situation looks like in terms of potential reward (number of handshakes) and what’s the best policy (the set of parameters) to use in order to maximise the reward value," says Margaritelli. "These are basically two sides of the same thing and by approaching this from these two ways the algorithm can converge quickly to very useful solutions."

Combined with a parasitic protocol running over Wi-Fi, the result is a virtual pet which learns how to maximise the number of handshakes captured over time — and which can communicate with other nearby Pwnagotchis in order to avoid duplicating effort, such as by splitting the channel list between units.

"Does it work? Yes it does," claims Margaritelli. "After a few days (or weeks, if you live in a isolated area), you literally start seeing the units going on different channels when they see each other, adjusting only to the channels where they 'see' potential reward, setting the timeouts correctly depending on how fast the unit is moving in space and therefore how fast it needs to 'lock on' new targets."

You needn't take Margaritelli's word for it, though: the development of the Pwnagotchi has been taking place in public, and it has developed an impressive following even before its full-version release — resulting in over 800 units registering for the PwnGRID distributed-Pwnagotchi network so far.

Margaritelli has a vision for the future, too: turning the PwnGRID into a distributed computing platform. "If and when the project will reach the thousands of units, PwnGRID will provide a similar amount of 'cores,' that can be orchestrated as a single computational unit, to everybody, for free," he explains. "Whatever cracking power the grid will reach, it’ll be distributed according to the previous contributions of who submitted the job: the more CPU cycles you’ll give to the grid, the higher the priority (and number of units) you will have to perform your operation."

Those interested in the Pwnagotchi project can find Margaritelli's full write-up on his website; the project itself has a dedicated website at pwnagotchi.ai which includes instructions on installation and configuration. The full source code, meanwhile, is available on the project's GitHub repository under the GNU General Public Licence Version 3.