Thread by @eredmil1

Sometimes as a security researcher I'm shocked that the assumptions we make: that every system can be broken, that no system is truly private are rarely held by the broader CS community.

A short thread of recent examples.

In preparing COVID19 ppr, I'm reminded of our efforts to submit the initial work on people's willingness to adopt based on privacy vs. accuracy @NeurIPSConf.

Reviewers claimed it was unnecessary to consider privacy in a decentralized system because it was perfectly private.

This is of course not true, as described by the creators of decentralized CT systems such as DP3T and as discovered by @AppCensusInc @v0max et al: themarkup.org/privacy/2021/0…

As security ppl, we took that breaches should simply be discussed in terms of likelihood, not impossibility.

Example 2: I spoke with a large industry AI lab last week.

One of their research heads said of our work on safety in digitally-mediated online-offline interaction like gig work & online dating:

"I don't see how that's about AI or how it's even research. That's product stuff"

A few years ago, comments like that derailed me.

I still sometimes say to mentors "but I'm not REALLY a CS researcher".

But most of the time, now, I shake my head because I know that they are wrong:

Security-thinking matters. Safety matters. And if we don't start considering security not just for data but for people before and while we build AI systems, we're doomed.

</end rant>