יום שני, 19 באוגוסט 2019

Exposure of TensorBoard interfaces in Fofa.so









Google AI family has an open source platform called Tensorflow, as part of Tensorflow there is a tool called TensorBoard.

"TensorFlow is an end-to-end open source platform for machine learning. It has a comprehensive, flexible ecosystem of tools, libraries and community resources that lets researchers push the state-of-the-art in ML and developers easily build and deploy ML powered applications"


This tool can be exposed to IoT search engines like Shodan, Fofa, Censys etc, as long as the users didn't properly secured the service.


I decided to use Fofa search engine for my proof of concept.


Attacker can search for TensorBoard in Fofa (similar to Shodan)






Most of the results I saw were not too sensitive 


Timeline:


17 Apr 2019
First email to Google security


After responding I gave them proper deadline.


14.05.2019
"The team is working on changing the behaviour in the next major release of TensorFlow."


16.05.2019
I frankly don't know their release plan :( They are now in Alpha if that helps: https://www.tensorflow.org/alpha


03.07.2019
you're more than welcome to publish, and we're happy to have a look too! :D


04.07.2019
It's unclear from the team, they've been informed you want to publish and to be honest, they don't seem overly concerned at the prospect of disclosure. I've asked them for a rough estimate of time for the release (which might take a couple of days for them to respond), but I'd say use your judgement on the best course of action here with the knowledge that the team says this was a pretty well documented and non-hidden aspect of Tensorboard. 


Sorry, I know that's really non-definitive and we do appreciate the patience and taking into account the fix here.. It's a weird situation, but I'll keep you up to date with anything the product team has to say about it
---==
Now there is TensorFlow 2.0 Beta out there so I decided to publish my findings.















אין תגובות:

הוסף רשומת תגובה