Kevin Bankston
@KevinBankston
Followers
11,881
Following
2,802
Media
514
Statuses
23,112
Senior Advisor on AI Governance @CenDemTech . Teaching AI law @GeorgetownLaw . Imagining better futures @ImaginationASU . My opinions, not theirs.
Washington, DC
Joined April 2008
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Still There
• 135593 Tweets
Arath
• 119674 Tweets
PROTECT BINI AT ALL COST
• 77834 Tweets
写真・動画
• 67975 Tweets
Demokratie
• 48635 Tweets
BUMP
• 31614 Tweets
#แค่น้องชายbyDMDBOYS
• 27089 Tweets
JKIA
• 25640 Tweets
MCM MEW IN SEOUL
• 23195 Tweets
牛角半額
• 22641 Tweets
PrabowoJKW RUKUNkompak
• 22577 Tweets
Ergebnis
• 22441 Tweets
#احسمو_صفقه_موليرو
• 22206 Tweets
BekerjaBERSATU UntukNKRI
• 21839 Tweets
食べ放題
• 20637 Tweets
Melt Down
• 16362 Tweets
DANGEROUS NEXT DOOR
• 14834 Tweets
#DANGEROUS_OUT_NOW
• 14793 Tweets
ASSDMS
• 10451 Tweets
Just pulled up my tax return in
@Google
Docs--and unbidden, Gemini summarized it. So...Gemini is automatically ingesting even the private docs I open in Google Docs? WTF, guys. I didn't ask for this. Now I have to go find new settings I was never told about to turn this crap off.
360
2K
24K
Ford's CEO just said on NPR that the future of profitability for the company is all the data from its 100 million vehicles (and the people in them) they'll be able to monetize. Capitalism & surveillance capitalism are becoming increasingly indistinguishable (and frightening).
197
3K
7K
It's also rather maddening to know that throwing more Gemini stuff in users' faces that they didn't ask for is one of the main reasons Google is completely blowing its climate goals as discussed just this morn on
@NPR
.
Thanks to
@darakerr
for talking with me about the high climate cost of generative AI, Google's reneging on its climate goals, and the real existential crisis facing us on NPR Morning Edition today.
9
112
308
5
112
1K
...what if I still don't want it looking at my docs unprompted? I didn't *ask* it to summarize my taxes, it just did. It should be up to me whether/which private docs prompt the model.
6
35
1K
Particularly with this sort of incident in its track record, Google should be giving us a lot more direct control over what if any of our content gets inputted into its models.
3
68
1K
A dangerous thing to say on multiple levels
Zuckerberg says platforms have responsibility for content now that technology can scan and classify whether it is "bad". Yikes
#Zuckerbowl
12
66
121
21
489
1K
So I asked Gemini to point me to the settings. This is what it said. Tried both, neither works. Gemini is *not* in Apps and services on my dashboard (1st option), and I didn't have a profile pic in the upper right of the Gemini page (2nd). Total fail,
@Google
. Little help?
8
35
1K
Garbage. Can’t wait to see how the next administration abuses this unwarranted expansion of a surveillance power we’ve already seen repeatedly abused.
President Joe Biden has signed the FISA Section 702 renewal bill into law.
24
94
126
42
356
1K
@Google
Notably, the only Gemini privacy I could find to manage was re: retention of my chats.
1
19
888
@MissMWilliams
I found the setting that’s supposed to keep Gemini out of my docs…and it was already turned off! Yet Gemini is absolutely giving me document summaries. A setting that doesn’t work is even worse than no setting.
6
58
862
"This is information that was public for most users..." Yes, AFTER Facebook forced all users to make it public in the 2009 privacy transition. Gee whiz, Mark.
3
275
655
Remember that time when David Duke, Julian Assange & Vladimir Putin all worked to put Donald Trump in the White House?
#whataworld
#memories
42
558
672
Once = anomaly, twice = pattern. After Shadow Brokers & Vault 7, it's clear: stockpiles of vulns eventually leak.
16
425
607
If you ever used Gemini with a particular PDF, it will re-invoke for every other PDF you load until you close it again. Same with GDocs--it wasn't on in any of my Docs, then I turned it on in one, and now it auto-summarizes any I open. Quite a sneaky sticky setting
@Google
!
2
51
674
Update: was pointed toward this page on the Gemini Workspace privacy commitments, saying that they don't use inputted data to train the underlying model. Great! But...
2
19
668
Want to stop info of yours from going out over Facebook's API? Then you'll need disable "Platform". (Go to privacy settings, and press "edit" for "apps, websites and plugins"). Go ahead and do it. I mean, who uses FB apps anyway? This ain't 2009.
22
468
643
Funny, doesn't look like Yahoo's transparency report indicated how EVERY SINGLE EMAIL was searched for the NSA.
9
550
599
And the model even claims to offer a setting that gives you that choice, yet that setting is AWOL at least in my account.
1
10
636
If this was a deliberate choice rather than some screwup, I can pretty clearly imagine the internal arguments that went into this very, very wrong
@Google
decision. So I'll just say: listen a bit more to your privacy and responsibility people and a bit less to your growth people.
3
18
610
...which again highlights why *it shoudn't ingest my documents without my prompting it to*
1
7
586
Notable looking at these privacy commitments how you have to pay an extra 20 bucks a month for the more privacy-protective AI Premium commitments. Otherwise, although the inputs still aren't used to train the underlying model they still use it for a bunch of other stuff...
2
26
562
Obama still hasn't taken a public stance against encryption backdoors. Tell him to speak up and help
#Savecrypto
!
http://t.co/hEVCpW61St
21
467
501
NSA scanning *content* of *all* int'l comms for info *about* targets is perhaps the most obviously unconstitutional thing it has been doing.
Now that NSA has stopped doing this "about" searching, Congress should go ahead & clarify that the law doesn't allow it in the first place!
4
18
21
18
314
473
If it ain't info or sympathy consider skipping it tonight. Save your analysis and agenda for tomorrow. Tonight, maybe just be a human being.
8
321
512
OK, more testing and I think I've figured it out (and it's still bad!). It seems that if you've ever clicked the Gemini button for a type of document then it remains open whenever you open another of that type--and therefore automatically ingests and summarizes it. So, e.g...
3
36
532
Dear
@TheJusticeDept
: have you submitted your iPhone exploit to
@WhiteHouse
's Vulnerability Equities Process yet? (Reporters, ask this q!)
9
385
472
Really disappointing thing about Yahoo building wiretap capability for NSA is they are in 9th circuit where law says they don't have to...
13
250
435
Makes one wonder: how many people have unwittingly inputted how many more private docs into Gemini simply because they clicked on that little AI star once in one document?
1
20
486
And none of this explains the case of the missing/hallucinated privacy settings.
3
14
453
Thank you
@missmwilliams
! The setting *does* exist (albeit not where Gemini said). But the punchline is in the next tweet…
0
5
66
4
30
443
Complain about cloud service privacy on X and the two most prevalent reply categories are “what did you expect, they are evil” vs “they already do x and y so why do you care about z,” both of which are basically nihilism. Thanks but I’m an advocate, not a nihilist. I want change.
7
13
440
funny not funny
maybe 538 can start a statistical model of which generals are most likely to side with trump regardless of the election outcome
10
140
501
6
55
409
3
1
425
Doing a little more testing--appears this is happening with any PDF of mine that I open from Drive. Thankfully not (yet?) automatically happening with Google Docs.
1
14
419
Thandie Newton’s exasperated answer at
#SXSW
to the question, “What does it feel like playing such a strong female character on
#Westworld
?”: “NORMAL!” (She also named her daughter after Ripley in Alien. Right on. ✊)
2
94
391
.
@Cloudflare
uses these 128 lava lamps as a 128-bit random number generator. Which is pretty, clever, and San Franciscan as all hell.
19
316
387
This continual "AI will save us from bad content" line is soooo dangerous.
11
90
360
@TenuredRadical
@Google
That is a fair question--I certainly wouldn't recommend it! But that doesn't absolve Google of responsibility here.
3
1
357
If you're wondering, the show was Freakonomics, and this story has some of the key transcript quotes.
3
102
305
Zuckerberg has now repeatedly said that Facebook will increasingly move toward AI automatically taking down and even *blocking* posts. Because that's what our policymakers want: in one breath they complain of Facebook's unchecked power and in another demand they use it even more.
11
110
297
Hey
@Ford
, you should note just how viral my previous tweet has gone. I don't have a huge following or anything. There's just a whole lotta drivers uncomfortable with this direction. They already paid for their cars with $$$, they don't also want to pay with their privacy.
8
41
253
Top 5 ISIS-recommended crypto apps are open source and/or non-US. Congress can only make *us* less secure, not them.
9
347
244
I'm very proud of the work--both internally, and from the many privacy advocates working on these issues--that has led to today's announcement that Facebook will be shutting down its facial recognition system and deleting over a billion face prints.
9
54
252
One of the fathers of open source is apparently a racist classist eugenicist, arguing that our meritocratic market has so efficiently grabbed all the "virtuous poor" & "deserving poor" that everyone left in the lower classes are increasingly genetically and socially irredeemable.
The dirty secret that confounds all modern attempts at improving outcomes by social engineering is that the low-IQ, the poor, and the badly behaved now pretty much coincide, statistically speaking.
Why? Because free markets and meritocracy mine the virtuous poor for people who
366
474
4K
27
56
232
Both
@Amazon
and
@Google
need to step up and own the fact that the steps they are taking right now on domain fronting will deprive vulnerable populations of secure services. They need too own it—and do something to fix it.
Amazon threatens to suspend Signal's AWS account over censorship circumvention:
115
3K
2K
10
123
217
Framing this as "Apple is blocking the cops" misses the point. The cops might feel like they are in an arms race with Apple. But Apple is actually in an arms race with the bad guys who can exploit the same vulnerabilities that the cops can.
7
117
205
I'm inspired by the bravery of those sharing their stories today to share mine: I lost my mom to suicide. I've always been ashamed to talk about it publicly. But that's not fair to her or anyone else suffering from mental illness. We need to
#stopthestigma
and talk about it.
9
22
210
Amazing:
@katecrawford
&
@trevorpaglen
and team have built a site where you can see how AI trained on the massive (and pretty racist and sexist and judge-y) ImageNet corpus categorizes you (or anyone else).
21
91
198
News: the amazing
@sarmorris
will be stepping up as
@OTI
’s director in June. I am leaving for a new job as Director of Privacy Policy at Facebook. Why, you might ask? The short answer: to do my best to make a difference at a critical time. The long answer:
53
44
206
I’m curious whether other people are showing the same (default?) toggles I’m seeing. Right now workplace and yt music are turned off but the others (yt video, flights, hotels, and maps (!!!) integrations are all on. Don’t think I’ve seen or touched these before…
2
4
205
2) Just because you invoke Gemini once in one doc doesn't mean you want it to then automatically ingest and summarize every doc you open of that type. That should *not* be a sticky setting.
1
4
203
…I certainly wouldn’t have turned off *music* while leaving *maps* integration on. Why are these set the way they are? Can’t see any logic to it.
3
3
196
4) (bonus problem) Again I ask: how many people have unwittingly had their private documents summarized by Gemini--which means that those summaries are all hanging out in your Gemini chat logs unless and until you delete them, unwittingly creating yet more privacy attack surface.
1
7
187
1) The Workplace integration opt-out (and potentially others?) doesn't appear to be working--I am and have been toggled off for Gemini in Workplace yet Gemini is still happily summarizing whatever doc I open.
1
4
184
3) Although the setting did exist, it wasn't where Gemini said. (I still don't know how to navigate to it; someone gave me a link.). Gemini needs to be trained at least to offer decent explanations of its own settings, and those settings need to be easy to find. They were not.
2
4
183
Also and just to be clear, gang: this isn't just about Ford. It's about every company that makes cars. And soon, pretty much every company that makes anything.
6
57
167
@flamesprite
@Google
I'm a little disturbed that you can't understand from my thread that I didn't intend to ask Gemini to do anything which is the entire frickin' point, while the one thing I did ask it to do, to see if it would give me an accurate answer about itself, gave me an inaccurate answer.
1
0
174
BREAKING: new report from House lawmakers rejects backdoors, says weakening encryption is against national interest:
3
156
167
Anyway. To sum up and hopefully close out for the eve, whether I set these settings or not--I'm eager to get screenshots of the Gemini signup flow I must have gone through at some point--I see at least 3 problems here:
1
2
173
Supreme Court holds 5-4 that the 4th amendment protects against unwarranted searches of cell phone location data. It took about 13 years of litigation--remember magistrate Smith and Orenstein's decisions in 2005?--but we finally got here.
6
73
156
Wait! Think I may have figured something out. In addition to being a Google One customer I also signed up for Workplace Labs last year to test the (then "Bard") features. I don't know what the flow for that signup had me agree to but I do see this. So...
1
7
165
Thu 9/28 in DC, experts discuss how surveillance disproportionately impacts people of color:
#colorofsurveillance
6
85
126
Wow—
@Facebook
just published a super-detailed guide to how they decide what content violates their guidelines, and now (finally!) offers ability to appeal a post’s takedown:
7
89
140
The White House "clarified" (lied) that this is an antifa symbol. Go ahead and Google for "antifa symbols". You'll see a whole lot of symbols that aren't an inverted red triangle. The *only* result with that symbol is from this very ad.
11
47
132
My working theory until Google can clarify is that using Workplace Labs overrode the explicit setting turning off Workplace integrations of Gemini (or, more likely chronology, I signed up for Workplace Labs and the later introduction of the controls didn't override that).
2
2
135
Alright. That's it for me tonight. So long and thanks for all the fish, see you around the school yard, etc. etc.
1
2
132
What a powerful and uniquely 21st century image: migrant workers seeking a signal to call home
http://t.co/QUFIze38gS
http://t.co/E3QwcXXM6m
3
195
119
The men's room at
#Dragoncon
. (I asked permission before taking the picture, I promise.)
http://t.co/AwnHF6Zo2P
3
96
121
Just so we're straight: PRISM (internet content). MARINA (internet metadata). NUCLEON (phone content). MAINWAY (phone metadata).
#NSA
8
238
116
Looking back, 2021 was the best year ever--despite everything!--because it's the year I married
@m_c_elish
, the most beautiful, loving, funny, whip-smart woman I've ever met, and my dearest friend. Looking forward to facing the new year with her, and all the years to come.
13
1
123
For all the career folk inside the federal government who are having to learn opsec to organize dissent. Stay safe, guys.
1
67
109
This is the correct take. This proposal would force "tech giants" to spend billions (fine, whatever) and *massively* over-censor (not fine, not whatever), while basically outlawing smaller competitors who literally can't afford it.
4
81
114
Autonomous pricing algorithms learn to collude to achieve near-monopoly pricing "by trial and error, with no prior knowledge of the environment in which they operate, without communicating with one another, and without being specifically designed or instructed to collude." (!!!)
Fascinating new research indicates that the AI behind online shopping prices may learn to collude! The independent, profit maximizing algorithms learned the tacit collusion strategy purely by trial & error -- and w/o any direct communication between them.
1
53
72
4
60
113
Wow: former FBI GC Jim Baker says the US government needs to "embrace reality" and support strong encryption or else put the nation at risk of "existential cybersecurity threats." Every hill staffer working this issue needs to read this. (So does AG Barr.)
2
74
106
Yes.
Did you ever think we would wake up and see headlines like these in the
@washingtonpost
and
@nytimes
!!
11K
6K
21K
5
23
102
An idea: use
@freedom
in locked mode to block access to your online accounts when crossing the border.
3
44
98
God bless
@VanJones68
for continually using his perch at CNN to fearlessly speak truth. Thanks, man.
0
19
96
Guys!
@doctorow
blogged about my my newly-published sci-fi story "Early Adopter" at
@BoingBoing
yesterday and called it "gorgeous and haunting." It's just a few words, but it's still the first positive review for my first ever short story. I am so chuffed!
7
11
95
Another irony: while Brazil rails against NSA spying, it attacks the encryption that can protect against NSA spying.
7
133
91
I find it horrifying that the first public meeting about this citywide sensor network happened *after* it was deployed. This is exactly why we need strong municipal laws requiring city council and citizen oversight and approval of these new technologies.
4
51
92
A great example of what we are worried about when it comes to automated takedowns of "extremist" material:
0
76
75
In 2001: A Space Odyssey, HAL 9000 read our lips. But as it turns out, our AI overlords will actually detect our monkey plots by watching micro-vibrations in our discarded bags of Frito-Lay snack products.
This algorithm can reconstruct an audio by analyzing minute vibrations of objects depicted in a video. In this experiment, intelligible speech was recovered from the vibrations of a potato-chip bag photographed from 15 feet away through soundproof glass
28
567
1K
6
39
89
Yet, as shown above, I interacted with it just today. So...this is another privacy feature that doesn't appear to be working correctly? And/or was overridden by signing up for Workplace Labs? Who knows. I gotta go to bed but hopefully will get more info tomorrow.
7
2
89
Sigh: it continues. I just went to the chat logs to delete the doc that was uploaded and the summary given in response. And...it's not there. According to this, I haven't used any Gemini features since last summer.
1
4
87
Google’s content moderation staff is bigger than the entire staffs of the vast majority of all internet companies. If this is the norm expected of global platforms, I hope you really like the platforms we have now because they are all we will ever have.
#HTLI
Google will have 10,000 people worldwide handling removal requests says Google chief litigator
#HTLI
0
21
20
3
57
80
If Apple can be ordered to do this, potentially any software vendor could be forced to update any device with malware. This is the fight.
3
125
74
I will note with some pride that when Facebook tried in 2009 to take away the ability to opt out of Platform API sharing, we at
@EFF
called them out hard (along with the Europeans, natch), and FB ultimately restored it. So please: use it!
1
31
76
“Cyberspace” and “metaverse” came from science fiction. Cryptonomicon was required reading at PayPal; Ready Player One at Oculus. What 21st century sci-fi should entrepreneurs read now for inspiration on what to build (or *not* build)?
25
22
79
Goodbye, DC. It’s been real. See you soon, California. Somehow I keep coming back to you.
4
0
80
13
135
74
Over at
@WIRED
,
@KimZetter
does a good job documenting and rebutting the post-Paris rhetoric against
#encryption
1
63
78
If for some reason you don't want to do that, at least edit "Apps Others Use" and uncheck all the categories of data there. (Note that data Facebook considers "public"--like name, gender, and friends list--will still be shared over API if you don't disable Platform.)
3
47
72
I am exactly the kind of nerd whose most expensive purchase on a romantic trip to Paris this weekend was...this 1st edition of Jules Verne’s “From Earth to The Moon” from 1865. (Suffice to say I have a loving partner who indulges my little passions.)
8
3
76