That's definitely a fair concern. We believe that open science and transparency are the right approach here. By releasing it, we ensure that everyone is on the same page with respect to abilities and defense.
Defending against such an "attack" is much easier if the technology is widely available and many people can play around with it and explore the limits.
At some point you have to let the rest of society in on it. As technology advances, if you keep the power away from the public, eventually you will split into two civilizations a vast divide creating an asymmetry that will be exploited by those on the elite side of the divide. It must be an aspect of technological rollout to figure out how to keep its usage safe once it is widespread.
You are assuming that such tech is not already being (or has been) developed covertly by malicious actors. Developing this and making it open source brings more awareness about the subject and will make it easier to develop defense models against such bots (whether already in existence or that will be developed in future).
> You are assuming that such tech is not already being (or has been) developed covertly by malicious actors.
I fail to see where playing out the "but others are doing it too" card exempts the responsibility of those who either lower or eliminate the barrier to entry to these attacks.
every AI sound/word/picture editor i've ran into says something along the lines of "we're releasing this data set to help stay secure in this day and age of easy counterfeiting of X.", but they never really mention how you apply the data in an adversarial way against itself -- they just sort of hand-wave that part.
Same with fake AI generated Obama video and sound, and earlier data-set generated chatbots; it's plastered all over the projects things like "Since these methods are available we think that it's important that this data is disseminated so that other's can use it to validate real world data sources", but again -- how?
We have the real data, we have the fake data -- how is this diff done, exactly?
I'm willing to bet it isn't as easy as all the AI researchers who release this stuff claim it may be.
If its secret or not publicly available people will argue using Occam’s razor or that only “State actors” could use this. With the subtext being your not important enough.
With the data public its more akin to driveby ssh login attempts. Not being important doesn’t mean your not under attack and people can take the necessary precautions.
That's a bit like saying that nuclear secrets should be made public so that people can "take precautions" because "anyone can have a nuclear weapon, not just state actors".
There are few reasonable ways to "take precautions" against nuclear weapons and there are few reasonable ways to "take precautions" against something like this short of swearing off of social media entirely.
Without reasonable defences, all you really accomplish is ramping up proliferation.
I don’t think weapons of mass destruction is comparable. More like a security vulnerability for the mind. You can no longer be sure its a human on the other side.