Softbank's 'Pepper' robot is a security joke

Softbank's popular anthropomorphic robot, Pepper, has myriad security holes according to research published by Scandinavian researchers earlier this month.

The 'bot allows unauthenticated root-level access, runs a Meltdown/Spectre-vulnerable processor, can be administered over unencrypted HTTP and has a default root password.

The researchers, Örebro University's Alberto Giaretta working with Michele De Donno and Nicola Drgoni of the Technical University of Denmark, decided that while there have been various one-off stunt-hacks of the robot, they weren't aware of any systematic assessment of Pepper's security.

Their research found that “it is a breeze to remotely turn it into a 'cyber and physical weapon', exposing malicious behaviours”.

A sorry list of vulns

The Meltdown/Spectre vulnerability was trivially identified, since all they needed was the uname -a terminal command.

A bit of Ettercap and Wireshark work revealed that the admin page is unsecured, which leaves naked the user/password pair of the only user account offered, called nao.

The robot almost implemented some kind of security by limiting SSH access to nao rather than permitting root SSH access, but since the root account's password is root (unchangeable and documented in the user manual), a superuser command from nao's terminal gives the attacker full privileges.

There's also no “log off” function in the admin panel.

“As a side note, we strongly believe that, in 2018, selling products so easily vulnerable to these kinds of attacks is not tolerable anymore”, the researchers observed.

The researchers also found the bot is susceptible to a brute-force attack.

This would most likely succeed, because Softbank's engineers haven't provided any protections against an attacker hammering Pepper with unlimited password attempts: “no countermeasures to brute-force attacks have been deployed with Pepper, which is once again an intolerable and disappointing finding”.

The researchers' next victim was an application on Pepper called 'Simple Animated Messages' (SAM), which lets users “design a simple choreography that makes Pepper move, say something through a text-to-speech service, and show a picture on its on-board tablet”.

By now you can guess what's coming: “the application performs no control over the file extension. As a matter of fact, we were able to upload images, text files which extensions have been modified to images, and even plain text files without performing extension editing”.

They decided not to build a proof-of-concept, but believe this would be easy for an attacker to exploit.

Then there's the Pepper API, which exposes its capabilities through languages like Python, C++, and Java, provides access to all sensors, cameras, microphones, and moving parts, and is “astonishingly insecure”.

“Pepper exposes a service on port 9559 which accepts TCP messages and reacts accordingly. As long as the messages comply to the API, by-design, Pepper accepts packets from whoever sends them” – and without authentication.

An attacker able to communicate with Pepper over TCP could use the cameras and microphones to spy on people and conversations; could remotely interact with people (for example, to trick them into providing personal information); use Pepper as an attack robot; or send the shutdown or factory reset commands.

At least Furby dolls needed hardware hacks to make them swear. ®