Charlie Savage has a story that confirms (he linked some of my earlier reporting) something I’ve long argued: NSA was willing to shut down the Internet dragnet in 2011 because it could do what it wanted using other authorities. In it, Savage points to an NSA IG Report on its purge of the PRTT data that he obtained via FOIA. The document includes four reasons the government shut the program down, just one of which was declassified (I’ll explain what is probably one of the still-classified reasons probably in a later post). It states that SPCMA and Section 702 can fulfill the requirements that the Internet dragnet was designed to meet. The government had made (and I had noted) a similar statement in a different FOIA for PRTT materials in 2014, though this passage makes it even more clear that SPCMA — DOD’s self-authorization to conduct analysis including US persons on data collected overseas — is what made the switch possible.
It’s actually clear there are several reasons why the current plan is better for the government than the previous dragnet, in ways that are instructive for the phone dragnet, both retrospectively for the USA F-ReDux debate and prospectively as hawks like Tom Cotton and Jeb Bush and Richard Burr try to resuscitate an expanded phone dragnet. Those are:
Both the domestic Internet and phone dragnet limited their use to counterterrorism. While I believe the Internet dragnet limits were not as stringent as the phone ones (at least in pre 2009 shutdown incarnation), they both required that the information only be disseminated for a counterterrorism purpose. The phone dragnet, at least, required someone sign off that’s why information from the dragnet was being disseminated.
Admittedly, when the FISC approved the use of the phone dragnet to target Iran, it was effectively authorizing its use for a counterproliferation purpose. But the government’s stated admissions — which are almost certainly not true — in the Shantia Hassanshahi case suggest the government would still pretend it was not using the phone dragnet for counterproliferation purposes. The government now claims it busted Iranian-American Hassanshahi for proliferating with Iran using a DEA database rather than the NSA one that technically would have permitted the search but not the dissemination, and yesterday Judge Rudolph Contreras ruled that was all kosher.
But as I noted in this SPCMA piece, the only requirement for accessing EO 12333 data to track Americans is a foreign intelligence purpose.
Additionally, in what would have been true from the start but was made clear in the roll-out, NSA could use this contact chaining for any foreign intelligence purpose. Unlike the PATRIOT-authorized dragnets, it wasn’t limited to al Qaeda and Iranian targets. NSA required only a valid foreign intelligence justification for using this data for analysis.
The primary new responsibility is the requirement:
- to enter a foreign intelligence (FI) justification for making a query or starting a chain,[emphasis original]
Now, I don’t know whether or not NSA rolled out this program because of problems with the phone and Internet dragnets. But one source of the phone dragnet problems, at least, is that NSA integrated the PATRIOT-collected data with the EO 12333 collected data and applied the protections for the latter authorities to both (particularly with regards to dissemination). NSA basically just dumped the PATRIOT-authorized data in with EO 12333 data and treated it as such. Rolling out SPCMA would allow NSA to use US person data in a dragnet that met the less-restrictive minimization procedures.
That means the government can do chaining under SPCMA for terrorism, counterproliferation, Chinese spying, cyber, or counter-narcotic purposes, among others. I would bet quite a lot of money that when the government “shut down” the DEA dragnet in 2013, they made access rules to SPCMA chaining still more liberal, which is great for the DEA because SPCMA did far more than the DEA dragnet anyway.
So one thing that happened with the Internet dragnet is that it had initial limits on purpose and who could access it. Along the way, NSA cheated those open, by arguing that people in different function areas (like drug trafficking and hacking) might need to help out on counterterrorism. By the end, though, NSA surely realized it loved this dragnet approach and wanted to apply it to all NSA’s functional areas. A key part of the FISC’s decision that such dragnets were appropriate is the special need posed by counterterrorism; while I think they might well buy off on drug trafficking and counterproliferation and hacking and Chinese spying as other special needs, they had not done so before.
The other thing that happened is that, starting in 2008, the government started putting FBI in a more central role in this process, meaning FBI’s promiscuous sharing rules would apply to anything FBI touched first. That came with two benefits. First, the FBI can do back door searches on 702 data (NSA’s ability to do so is much more limited), and it does so even at the assessment level. This basically puts data collected under the guise of foreign intelligence at the fingertips of FBI Agents even when they’re just searching for informants or doing other pre-investigative things.
In addition, the minimization procedures permit the FBI (and CIA) to copy entire metadata databases.
FBI can “transfer some or all such metadata to other FBI electronic and data storage systems,” which seems to broaden access to it still further.
Users authorized to access FBI electronic and data storage systems that contain “metadata” may query such systems to find, extract, and analyze “metadata” pertaining to communications. The FBI may also use such metadata to analyze communications and may upload or transfer some or all such metadata to other FBI electronic and data storage systems for authorized foreign intelligence or law enforcement purposes.
In this same passage, the definition of metadata is curious.
For purposes of these procedures, “metadata” is dialing, routing, addressing, or signaling information associated with a communication, but does not include information concerning the substance, purport, or meaning of the communication.
I assume this uses the very broad definition John Bates rubber stamped in 2010, which included some kinds of content. Furthermore, the SMPs elsewhere tell us they’re pulling photographs (and, presumably, videos and the like). All those will also have metadata which, so long as it is not the meaning of a communication, presumably could be tracked as well (and I’m very curious whether FBI treats location data as metadata as well).
Whereas under the old Internet dragnet the data had to stay at NSA, this basically lets FBI copy entire swaths of metadata and integrate it into their existing databases. And, as noted, the definition of metadata may well be broader than even the broadened categories approved by John Bates in 2010 when he restarted the dragnet.
So one big improvement between the old domestic Internet dragnet and SPCMA (and 702 to a lesser degree, and I of course, improvement from a dragnet-loving perspective) is that the government can use it for any foreign intelligence purpose.
At several times during the USA F-ReDux debate, surveillance hawks tried to use the “reform” to expand the acceptable uses of the dragnet. I believe controls on the new system will be looser (especially with regards to emergency searches), but it is, ostensibly at least, limited to counterterrorism.
One way USA F-ReDux will be far more liberal, however, is in dissemination. It’s quite clear that the data returned from queries will go (at least) to FBI, as well as NSA, which means FBI will serve as a means to disseminate it promiscuously from there.
Another thing replacing the Internet dragnet with 702 access does it provide another way to correlate multiple identities, which is critically important when you’re trying to map networks and track all the communication happening within one. Under 702, the government can obtain not just Internet “call records” and the content of that Internet communication from providers, but also the kinds of thing they would obtain with a subpoena (and probably far more). As I’ve shown, here are the kinds of things you’d almost certainly get from Google (because that’s what you get with a few subpoenas) under 702 that you’d have to correlate using algorithms under the old Internet dragnet.
Every single one of these data points provides a potentially new identity that the government can track on, whereas the old dragnet might only provide an email and IP address associated with one communication. The NSA has a great deal of ability to correlate those individual identifiers, but — as I suspect the Paris attack probably shows — that process can be thwarted somewhat by very good operational security (and by using providers, like Telegram, that won’t be as accessible to NSA collection).
This is an area where the new phone dragnet will be significantly better than the existing phone dragnet, which returns IMSI, IMEI, phone number, and a few other identifiers. But under the new system, providers will be asked to identify “connected” identities, which has some limits, but will nonetheless pull some of the same kind of data that would come back in a subpoena.
While replacing the domestic Internet dragnet with SPCMA provides additional data with which to do correlations, much of that might fall under the category of additional functionality. There are two obvious things that distinguish the old Internet dragnet from what NSA can do under SPCMA, though really the possibilities are endless.
The first of those is content scraping. As the Intercept recently described in a piece on the breathtaking extent of metadata collection, the NSA (and GCHQ) will scrape content for metadata, in addition to collecting metadata directly in transit. This will get you to different kinds of connection data. And particularly in the wake of John Bates’ October 3, 2011 opinion on upstream collection, doing so as part of a domestic dragnet would be prohibitive.
In addition, it’s clear that at least some of the experimental implementations on geolocation incorporated SPCMA data.
I’m particularly interested that one of NSA’s pilot co-traveler programs, CHALKFUN, works with SPCMA.
Chalkfun’s Co-Travel analytic computes the date, time, and network location of a mobile phone over a given time period, and then looks for other mobile phones that were seen in the same network locations around a one hour time window. When a selector was seen at the same location (e.g., VLR) during the time window, the algorithm will reduce processing time by choosing a few events to match over the time period. Chalkfun is SPCMA enabled1.
1 (S//SI//REL) SPCMA enables the analytic to chain “from,” “through,” or “to” communications metadata fields without regard to the nationality or location of the communicants, and users may view those same communications metadata fields in an unmasked form. [my emphasis]
Now, aside from what this says about the dragnet database generally (because this makes it clear there is location data in the EO 12333 data available under SPCMA, though that was already clear), it makes it clear there is a way to geolocate US persons — because the entire point of SPCMA is to be able to analyze data including US persons, without even any limits on their location (meaning they could be in the US).
That means, in addition to tracking who emails and talks with whom, SPCMA has permitted (and probably still does) permit NSA to track who is traveling with whom using location data.
Finally, one thing we know SPCMA allows is tracking on cookies. I’m of mixed opinion on whether the domestic Internet ever permitted this, but tracking cookies is not only nice for understanding someone’s browsing history, it’s probably critical for tracking who is hanging out in Internet forums, which is obviously key (or at least used to be) to tracking aspiring terrorists.
Most of these things shouldn’t be available via the new phone dragnet — indeed, the House explicitly prohibited not just the return of location data, but the use of it by providers to do analysis to find new identifiers (though that is something AT&T does now under Hemisphere). But I would suspect NSA either already plans or will decide to use things like Supercookies in the years ahead, and that’s clearly something Verizon, at least, does keep in the course of doing business.
All of which is to say it’s not just that the domestic Internet dragnet wasn’t all that useful in its current form (which is also true of the phone dragnet in its current form now), it’s also that the alternatives provided far more than the domestic Internet did.
Jim Comey recently said he expects to get more information under the new dragnet — and the apparent addition of another provider already suggests that the government will get more kinds of data (including all cell calls) from more kinds of providers (including VOIP). But there are also probably some functionalities that will work far better under the new system. When the hawks say they want a return of the dragnet, they actually want both things: mandates on providers to obtain richer data, but also the inclusion of all Americans.
Many many many outlets are reporting that China has continued conducting economic espionage even after Xi Jinping agreed to stop doing it. They base that claim on this post from CloudStrike, a big cybersecurity contractor that spends a lot of time feeding the press scary stories about hacking.
Here’s the proof they offer:
Over the last three weeks, CrowdStrike Falcon platform has detected and prevented a number of intrusions into our customers’ systems from actors we have affiliated with the Chinese government. Seven of the companies are firms in the Technology or Pharmaceuticals sectors, where the primary benefit of the intrusions seems clearly aligned to facilitate theft of intellectual property and trade secrets, rather than to conduct traditional national-security related intelligence collection which the Cyber agreement does not prohibit.
In addition to preventing these intrusions, the CrowdStrike Falcon platform also provided full visibility into every tool, command and technique used by the adversary. This allowed us to determine that the hackers saw no need to change their usual tradecraft or previously used infrastructure in an attempt to throw off their scent.
The include a timeline showing 9 attempted intrusions into Tech Sector companies, and 2 into Pharma companies since Xi and President Obama signed the hacking agreement.
Now, even assuming that CrowdStrike has accurately labeled these Chinese government hackers (CrowdStrike’s CTO was less confident in an interview with Motherboard) this still is not proof that China has violated the agreement.
After all, the key part of the agreement is on how stolen information gets used — whether it gets used to benefit individual companies or even entire sectors (the latter of which we do in our own spying, but never mind). If CrowdStrike prevented any data from being stolen, then it is impossible to assert that it was being stolen to benefit market actors without more evidence that the hackers were tasked by a market actor. Even the indictment everyone points to as proof that China engages in economic espionage did not allege that the People Liberation’s Army had shared the data involved in the single economic espionage charge with private sector companies, and given that the data in question pertained to nuclear technology ,it’s not something that is proven just because it was stolen in the context of an ongoing relationship with the victim (even if that is a logical presumption to make).
The same is true here. When China hacked Google to spy on dissidents, that was clearly national security spying. When the US hacked Huawei to figure out how to backdoor its equipment, that was clearly national security spying.When the US used Microsoft and Siemens products to carry out StuxNet, the tech companies were merely enabling targets. There are too many reasons to hack tech sector companies for solidly national security purposes to claim, just based on the sector itself, that it was done for economic espionage.
You can’t even point to the 2 Pharma intrusions to make the claim. A list of sites the State Department identified as critical infrastructure from a leaked 2009 cable includes over 25 pharmaceutical sites (including animal Pharma), many of them related to vaccines. If we’re treating pharmaceutical supply and research facilities as critical infrastructure, with the presumed consequent defensive surveillance of those sites, it is tough to argue the Chinese can’t consider our pharmaceutical companies making key drugs to be critical targets. Both can be argued to stem from the same public health concerns.
I’m not saying it’s impossible or even unlikely that these intrusions were attempted economic espionage. I’m saying that this isn’t evidence of it, and that the reporting repeating this claim has been far too credulous.
But that also points to one of the inherent problems with this deal (one pointed to by many people at the time). When last he testified on the subject, Jim Clapper didn’t even claim to have fully attributed the OPM hack. The same attribution and use problems exist here. China may steal data on an important new drug, but that’s not going to be enough to prove they stole it for commercial gain until they release their own copycat of the drug in several years and use it to undercut the US company’s product, and even then that may require a lot more data — collected by spying! — from inside the market companies themselves (in part because China engages in many other means of stealing data which aren’t the subject of a special agreement, which will make even the copycat instance hard to prove came from an intrusion).
China knew that, too, when it signed the agreement. It will take more than evidence of 11 attempted intrusions to prove that China is violating the agreement.
The Computer and Communications Industry Association — a trade organization that represents Internet, social media, and even some telecom companies — came out yesterday against the Cyber Intelligence Sharing Act, an information sharing bill that not only wouldn’t be very useful in protecting against hacking, but might have really dangerous unintended consequences, such as gutting regulatory authority over network security negligence (though the Chamber of Commerce, this bill’s biggest backer, may not consider it an unintended consequence).
Most coverage of this decision emphasizes CCIA’s concern about the bill’s danger to privacy.
CISA’s prescribed mechanism for sharing of cyber threat information does not sufficiently protect users’ privacy or appropriately limit the permissible uses of information shared with the government.
But I’m far more interested in CCIA’s stated concern that the bill, in authorizing defensive measures, would permit actions that would damage the Internet’s infrastructure (to which a number of these companies contribute).
In addition, the bill authorizes entities to employ network defense measures that might cause collateral harm to the systems of innocent third parties.
But such a system … must not enable activities that might actively destabilize the infrastructure the bill aims to protect.
At least some of these companies that make up our Internet ecosystem think that some other companies, in aggressively pursuing perceived intruders to their systems, will do real damage to Internet as a whole.
It seems like a worthy concern. And yet the Senate runs headlong towards passing this bill anyway.
Wired’s hack-of-the-day story reports that researchers hacked a Tesla (unlike the Chrysler hack, it required access to the vehicle once, though the Tesla also has a browser vulnerability that might not require direct access).
Two researchers have found that they could plug their laptop into a network cable behind a Model S’ driver’s-side dashboard, start the car with a software command, and drive it. They could also plant a remote-access Trojan on the Model S’ network while they had physical access, then later remotely cut its engine while someone else was driving.
The story notes how much more proactive Tesla was in patching this problem than Chrysler was.
The researchers found six vulnerabilities in the Tesla car and worked with the company for several weeks to develop fixes for some of them. Tesla distributed a patch to every Model S on the road on Wednesday. Unlike Fiat Chrysler, which recently had to issue a recall for 1.4 million cars and mail updates to users on a USB stick to fix vulnerabilities found in its cars, Tesla has the ability to quickly and remotely deliver software updates to its vehicles. Car owners only have to click “yes” when they see a prompt asking if they want to install the upgrade.
In my understanding, Tesla was able to do this both because it responded right away to implement the fix, and because it had the technical ability to distribute the update in such a way that was usable for end users. Chrysler deserves criticism for the former (though at least according to Chrysler, it did start to work on a fix right away, it just didn’t implement it), but the latter is a problem that will take some effort to fix.
Which is one reason I think a better comparison with Tesla’s quick fix is Google’s delayed fix for the Stagefright vulnerability. As the researcher who found it explained, Google address the vulnerability internally immediately, just like Tesla did.
Google has moved quickly to reassure Android users following the announcement of a number of serious vulnerabilities.
The Google Stagefright Media Playback Engine Multiple Remote Code Execution Vulnerabilitiesallow an attacker to send a media file over a MMS message targeting the device’s media playback engine, Stagefright, which is responsible for processing several popular media formats.
Attackers can steal data from infected phones, as well as hijacking the microphone and camera.
Android is currently the most popular mobile operating system in the world — meaning that hundreds of millions of people with a smartphone running Android 2.2 or newer could be at risk.
Joshua Drake, mobile security expert with Zimperium, reports
A fully weaponized successful attack could even delete the message before you see it. You will only see the notification…Unlike spear-phishing, where the victim needs to open a PDF file or a link sent by the attacker, this vulnerability can be triggered while you sleep. Before you wake up, the attacker will remove any signs of the device being compromised and you will continue your day as usual – with a trojaned phone.
Zimperium say that “Google acted promptly and applied the patches to internal code branches within 48 hours, but unfortunately that’s only the beginning of what will be a very lengthy process of update deployment.”
But with Android the updates need to go through manufacturers, which creates a delay — especially given fairly crummy updating regimes by a number of top manufacturers.
The experience with this particular vulnerability may finally be pushing Android-based manufacturers to fix their update process.
It’s been 10 days since Zimperium’s Joshua Drake revealed a new Android vulnerabilitycalled Stagefright — and Android is just starting to recover. The bug allows an attacker to remotely execute code through a phony multimedia text message, in many cases without the user even seeing the message itself. Google has had months to write a patch and already had one ready when the bug was announced, but as expected, getting the patch through manufacturers and carriers was complicated and difficult.
But then, something unexpected happened: the much-maligned Android update system started to work. Samsung, HTC, LG, Sony and Android One have already announced pending patches for the bug, along with a device-specific patch for the Alcatel Idol 3. In Samsung’s case, the shift has kicked off an aggressive new security policy that will deploy patches month by month, an example that’s expected to inspire other manufacturers to follow suit. Google has announced a similar program for its own Nexus phones. Stagefright seems to have scared manufacturers and carriers into action, and as it turns out, this fragmented ecosystem still has lots of ways to protect itself.
I make this comparison for two reasons. One, if Google — the customers of which have the hypothetical ability to send out remote patches, even if they’ve long neglected that ability — still doesn’t have this fixed, it’s unsurprising that Chrysler doesn’t yet.
But some of the additional challenges that Chrysler has that Tesla has fewer of stem from the fragmented industry. Chrysler’s own timeline of its vulnerability describes a “third party” discovering the vulnerability (not the hackers), and a “supplier” fixing it.
In January 2014, through a penetration test conducted by a third party, FCA US LLC (“FCA US”) identified a potential security vulnerability pertaining to certain vehicles equipped with RA3 or RA4 radios.
A communications port was unintentionally left in an open condition allowing it to listen to and accept commands from unauthenticated sources. Additionally, the radio firewall rules were widely open by default which allowed external devices to communicate with the radio. To date, no instances related to this vulnerability have been reported or observed, except in a research setting.
The supplier began to work on security improvements immediately after the penetration testing results were known in January 2014.
But it’s completely unclear whether that “third party” is the “supplier” in question. Which means it’s unclear whether this was found in the supplier’s normal testing process or in something else.
One reason cars are particularly difficult to test are because so many different suppliers provide parts which don’t get tested (or even adequately specced) in an integrated fashion.
Then, if you need to fix something you can’t send out over a satellite or Internet network, you’re dealing with the — in many cases — archaic relationships car makers have with dealers, not to mention the limitations of dealer staff and equipment to make the fix.
I don’t mean to excuse the automotive industry — they’re going to have to fix these problems (and the same problems lie behind fixing some of the defects tied to code that doesn’t stem from hacks, too, such as Toyota’s sudden acceleration problem).
It’s worth noting, however, how simplified supply and delivery chains make fixing a problem a lot easier for Tesla than it is for a number of other entities, both in and outside of the tech industry.
UPDATE — 4:30 PM EDT —
Hey, it’s Rayne here, adding my countervailing two cents (bitcoins?) to the topic after Marcy and I exchanged a few emails about this topic. I have a slightly different take on the situation since I’ve done competitive intelligence work in software, including open source models like Android.
Comparing Fiat Chrysler’s and Google’s Android risks, the size and scale of the exposures are a hell of a lot different. There are far more Android devices exposed than Chrysler car models at risk — +1 billion Android devices shipped annually around the globe as of 4Q2014.
Hell, daily activations of Android devices in 2013 were 1.2 million devices per day — roughly the same number as all the exposed Chrysler vehicles on the road, subject to recall.
Google should have a much greater sense of urgency here due to the size of the problem.
Yet chances of a malware attack on an Android device actually causing immediate mortal threat to one or more persons is very low, compared to severity of Chrysler hack. Could a hacker tinker with household appliances attached via Android? It’s possible — but any outcome now is very different from a hacker taking over and shutting down a vehicle operating at high speed in heavy traffic, versus shutting off a Phillips remote-controlled Hue lamp or a Google Nest thermostat, operating in the Internet of Things. The disparity in annoyance versus potential lethality may explain why Google hasn’t acted as fast as Tesla — but it doesn’t explain at all why Chrysler didn’t handle announcing their vulnerability differently. Why did they wait nearly a year to discuss it in public? Continue reading
This morning, Wired reports that the hackers who two years ago hacked an Escape and a Prius via physical access have hacked a Jeep Cherokee via remote (mobile phone) access. They accessed the vehicle’s Electronic Control Unit and from that were able to get to ECUs controlling the transmission and brakes, as well as a number of less critical items. The hackers are releasing a report [correction: this is Markey’s report], page 86 of which explains why cars have gotten so much more vulnerable (generally, a combination of being accessible via external communication networks, having more internal networks, and having far more ECUs that might have a vulnerability). It includes a list of the most and least hackable cars among the 14 they reviewed.
Today Ed Markey and Richard Blumenthal are releasing a bill meant to address some of these security vulnerabilities in cars.
Meanwhile — in a remarkably poorly timed announcement — Apple announced yesterday that it had hired Fiat Chrysler’s former quality guy, the guy who would have overseen development of both the hackable Jeep Cherokee and the safer Dodge Viper.
Doug Betts, who led global quality at Fiat Chrysler Automobiles NV until last year, is now working for the Cupertino, Calif.-based electronics giant but declined to comment on the position when reached Monday. Mr. Betts’ LinkedIn profile says he joined Apple in July and describes his title as “Operations-Apple Inc.” with a location in the San Francisco Bay Area but no further specifics.
Along with Mr. Betts, whose expertise points to a desire to know how to build a car, Apple recently recruited one of the leading autonomous-vehicle researchers in Europe and is building a team to work on those systems.
In 2009, when Fiat SpA took over Chrysler, CEO Sergio Marchionne tapped Mr. Betts to lead the company’s quality turnaround, giving him far-reaching authority over the company’s brands and even the final say on key production launches.
Mr. Betts abruptly left Fiat Chrysler last year to pursue other interests. The move came less than a day after the car maker’s brands ranked poorly in an influential reliability study.
Note, the poor quality ratings that preceded Betts’ departure from Fiat Chrysler pertained especially to infotainment systems, which points to electronics vulnerabilities generally.
As they get into the auto business, Apple and Google will have the luxury that struggling combustion engine companies don’t have — that they’re not limited by tight margins as they try to introduce bells and whistles to compete on the marketplace. But they’d do well to get this quality and security issue right from the start, because the kind of errors tech companies can tolerate — largely because they can remotely fix bugs and because an iPhone that prioritized design over engineering can’t kill you — will produce much bigger problems in cars (though remote patching will be easier in electric cars).
So let’s hope Apple’s new employee takes this hacking report seriously.
I’m going to make an unpopular argument.
Most observers of USA F-ReDux point to weakened transparency provisions as one of the biggest drawbacks of the latest version of the bill. They’re not wrong: transparency procedures are worse, remarkably so.
But given that I already thought they were not only inadequate but dangerously misleading,* I’m actually grateful to have had the Intelligence Community do another version of transparency provisions, which shows what they’re most intent on hiding and/or hints at what they will really be doing behind the carefully scripted words they’re getting Congress to rubber-stamp.
The most remarkable of the changes in the transparency provision is that they basically took out this language requiring a top level count of Section 702 targets and persons whose communications were affected — this language.
(i) the number of targets of such orders;
(ii) the number of individuals whose communications were collected pursuant to such orders; [sub 500 range]
(iii) the number of individuals whose communications were collected pursuant to such orders who are reasonably believed to have been located in the United States at the time of collection; [sub 500 range]
This leaves — in addition to the “number of 702 orders” requirement — just this reporting requirement for back door content and metadata searches which (like the Leahy bill) exempts the gross majority of the back door searches, because they are done by the FBI.
(A) the number of search terms concerning a known United States person used to retrieve the unminimized contents of electronic communications or wire communications obtained through acquisitions authorized under such section, excluding the number of search terms used to prevent the return of information concerning a United States person; and [FBI Exemption]
(B) the number of queries concerning a known United States person of unminimized noncontents information relating to electronic communications or wire communications obtained through acquisitions authorized under such section, excluding the number of queries containing information used to prevent the return of information concerning a United States person; [FBI Exemption]
In other words, ODNI was happy to tell us that the number of FISA 702 targets went up by 4% between 2013 and 2014, but not how much those numbers of targets will go up in 2015, when they presumably begin to roll out the new call chaining provision.
I suspect — and these are well educated but nevertheless wildarseguesses — there are several reasons.
First, the reporting provisions as a whole move from tracking “individuals whose communications were collected” to “unique identifiers used to communicate information.” They probably did that because they don’t really have a handle on which of the identifiers all represent the same natural person (and some aren’t natural persons), and don’t plan on ever getting a handle on that number. Under last year’s bill, ONDI could certify to Congress that he couldn’t count that number (and then as an interim measure I understand they were going to let them do that, but require a deadline on when they would be able to count it). Now, they’ve eliminated such certification for all but 702 metadata back door searches (that certification will apply exclusively to CIA, since FBI is exempted). In other words, part of this is just an admission that ODNI does not know and does not planning on knowing how many of the identifiers they target actually fit together to individual targets.
But since they’re breaking things out into identifiers now, I suspect they’re unwilling to give that number because for each of the 93,000 targets they’re currently collecting on, they’re probably collecting on at least 10 unique identifiers and probably usually far, far more.
Just as an example (this is an inapt case because Hassanshahi, as a US person, could not be a PRISM target, but it does show the bare minimum of what a PRISM target would get), the two reports Google provided in response to administrative subpoenas for information on Shantia Hassanshahi, the guy caught using the DEA phone dragnet (these were subpoenas almost certainly used to parallel construct data obtained from the DEA phone dragnet and PRISM targeted at the Iranian, “Sheikhi,” they found him through), included:
So just for this person who might be targeted under the new phone dragnet (though they’d have to play the same game of treating Iran as a terrorist organization that they currently do, but I assume they will), you’d have upwards of 15 unique identifiers obtained just from Google. And that doesn’t include a single cookie, which I’ve seen other subpoenas to Google return.
In other words, one likely reason the IC has decided, now that they’re going to report in terms of unique identifiers, they can’t report the number of identifiers targeted under PRISM is because it would make it clear that those 93,000 targets represent, very conservatively, over a million identifiers — and once you add in cookies, maybe a billion identifiers — targeted. And reporting that would make it clear what kind of identifier soup the IC is swimming in.
There is another reason I think they’ve grown reluctant to show much transparency under 702. Implementing the USA F-ReDux system — in which each provider sets up facilities they can use to chain on non-call detail record session identifying information — means more providers (smaller phone companies, and some new Internet providers, for example) will have what amount to PRISM-lite portals that can also be used for PRISM production. If you build it they will come!
In addition, Verizon and Sprint may be providing more PRISM smart phone materials in addition to upstream collection (AT&T likely already provides a lot of this because that’s how they roll).
So I suspect that, whereas now there’s a gap between the cumulative numbers providers report in their own transparency reports and what we see from ODNI, that number will grow notably, which would lead to questions about where the additional 702 production was coming from. (Until Amazon starts producing transparency reports, though, I’ll just assume they’re providing it all).
Finally, I think that once USA F-ReDux rolls out, the government (read, FBI, where this data will first be sucked in) will have difficulty distinguishing between the 702 and 215 production from a number of providers — probably AT&T, Verizon, Apple, Google, and Microsoft, but that’s just a guess.
Going back to the case of Hassanshahi, for example (and assuming, as I do, that the government has been parallel constructing the fact that they also targeted the Iranian Sheikhi identifier under PRISM, which would have immediately led them to his GMail account, as they very very easily could), the Tehran phone to Google call between Sheikhi and Hassanshahi would likely come in via at least 3 sources: Sheihki PRISM collection, Google USA F-ReDux returns on the Sheikhi number, and AT&T backbone USA F-ReDux returns on the Sheikhi number. And all that’s before you’ve taken a single hop into Hassanshahi’s accounts.
In other words, what you’re actually getting with USA F-ReDux is a way to get to the metadata of US persons identified via incidental collection under PRISM (again, this should just before for targets of a somewhat loosey goosey definition of terrorism targets). It’s basically a way to get a metadata “hop” off of all the Americans already “incidentally” collected under PRISM (note, permission to do this for targets identified under a probable cause warrant is already written into every phone dragnet order; this just extends that, with FISC review, to PRISM targets). And for the big providers that have anything that might be considered “call” service, the portals from which that will derive will likely be very very closely related.
Thus far, none of the Internet providers who have issued statements in support of the latest incarnation of USA Freedom Act (which I’m calling USA F-ReDux) have mentioned that they will be getting expansive immunity and compensation for helping the government spy on you.
Along with two other features, Google argues USA F-ReDux would,
[E]nd the bulk collection of communications metadata under various legal authorities. This not only includes telephony metadata collected under Section 215, but also Internet metadata that has been or could be collected under other legal authorities.
I find that an interesting way to describe the bill, particularly given that Google calls this “modernizing” surveillance, not limiting it.
Congress Has Only A Few Weeks Left to Modernize Surveillance Laws
Both the government and some providers used that same language — “modernize” — during the FISA Amendments Act, too. Sure, that was partly because it accommodated the law to growing Internet reliance. USA F-ReDux will do that too, to the extent it allows the government to obtain metadata for things like Google Meet-Ups and other VOIP calls and Internet messaging, which the government needs if it really wants dragnet coverage. FAA also involved deputizing Internet providers so that their data could not longer be collected in bulk by phone companies.
Modernizing surveillance, they called that.
And as I’ve just begun to lay out, this bill will set up a system similar in many respects to PRISM, where the government would go to the provider to get what they wanted on a target. Under PRISM, what the government wanted quickly expanded. Within 6 months of the roll-out of PRISM, the government was already asking for 9 different types of data from providers like Yahoo, apparently spanning Yahoo’s four business functions (meaning email, information services, data storage, and Yahoo internal functions).
Here, as with FAA, the government will go to providers to get what they want. And given that the bill permits the government to ask providers to chain on non-Call Detail Record session identifiers (things like cookies and location data), the government will benefit from, though not directly access, some of the same data that the government started obtaining under PRISM. And while I would hope the FISA Court would exert some oversight, I would also bet the government will make increasingly expansive claims about what constitutes a “session identifier” that can be used to chain (we know that, overseas, they chain on address books and photographs, for example).
And in one way, USA F-ReDux is worse than PRISM. Unlike FAA, USA F-ReDux will feature an added role for a Booz-type contractor compiling all this data, possibly in some cloud somewhere that would be about as safe as all the documents Edward Snowden took, to make it easier to chain across providers.
This is what Google celebrates as “modernization.”
But let’s go back to Google’s representation of this as ending bulk collection of, “Internet metadata that has been or could be collected under other legal authorities.”
We’ve long discussed the Section 215 dragnet as covering just calls made by phone companies (though Verizon’s Counsel, in a hearing last year, noted that the government would have to get VOIP if it wanted full coverage).
But that’s not true. As I reported the other day, at least one of the phone metadata dragnets was collecting VOIP metadata. Google’s VOIP metadata. In fact, the only known use of the DEA dragnet involved a US user subscribing to Google calls.
In other words, the Shantia Hassanshahi case is important not just because it led to us learning about the DEA dragnet, but because it revealed that (in addition to Google’s Internet metadata being collected under PRTT illegally for years), Google’s VOIP data also got sucked up in at least one phone dragnet.
Google doesn’t like other people being able to spy on its customers.
But now that USA F-ReDux will return it to the position of having the monopoly on spying on its customers, it calls this “modernization.”
As I noted in this post, the government insists that it did not engage in parallel construction in the case of Shantia Hassanshahi, the Iranian-American busted for sanctions violations using evidence derivative of a search of what the government now claims was a DEA dragnet. “While it would not be improper for a law enforcement agency to take steps to protect the confidentiality of a law enforcement sensitive investigative technique, this case raises no such issue.”
The claim is almost certainly bullshit, true in only the narrowest sense.
Indeed, the changing story the government has offered about how they IDed Hassanshahi based off a single call he had with a phone belonging to a person of interest, “Sheikhi,” in Iran, is instructive not just against the background of the slow reveal of multiple dragnets over the same period. But also for the technological capabilities included in those claims. Basically, the government appears to be claiming they got a VOIP call from a telephony database.
As I lay out below, the story told by the government in various affidavits and declarations (curiously, the version of the first one that appears in the docket is not signed) changed in multiple ways. While there were other changes, the changes I’m most interested in pertain to:
As you can see from the excerpts below, Akronowitz at first claimed to have searched “HSI-accessible law enforcement databases,” plural, and suggested he searched them himself. In July 2014, in response to a motion to suppress (and after Edward Snowden had disclosed the NSA’s phone dragnet), Akronowitz changed that story and said he sent a research request to a single database, implying someone else did a search of just one database. Akronowitz told the same story in yet another revised affidavit submitted last October. In the declaration submitted in December but unsealed in January, DEA Assistant Special Agent Robert Patterson stuck with the single database story and used the passive voice to hide who did the database query.
While Akronowitz’ story didn’t change regarding how he discovered that Hassanshahi’s phone was a Google number, it did get more detailed in the July 2014 affidavit, which explained that he had first checked with another VOIP provider before being referred to Google.
Perhaps most interestingly, the government’s story changed regarding how many calls of interest there were, and between what numbers. In January 2013, Akronowitz said “a number of telephone calls between ‘Sheikhi’s’ known business telephone number and telephone number 818-971-9512 had occurred within a relatively narrow time frame” (though he doesn’t tell us what that time frame was). He also says that his Google subpoena showed “numerous calls to the same Iranian-based telephone number during a relatively finite period of time.” He neither explained that this number was not Sheikhi’s number — it was a different Iranian number — nor what he means by “a relatively finite period of time.” His July and October affidavits said his research showed a contact, “on one occasion, that is, on July 4, 2011,” with Sheikhi’s number. The July affidavit maintained the claim that there were multiple calls between Hassanshahi’s number and an Iranian one: “numerous phone calls between Hassanshahi’s ‘818’ number and one Iranian phone number.” But by October, Akronowitz conceded that the Google records showed only “that Hassanshahi’s ‘818’ number made contact with an Iranian phone number (982144406457) only once, on October 5, 2011” (as well as a “22932293” number that he bizarrely claimed was a call to Iran). Note, Akronowitz’ currently operative story would mean the government never checked whether there were any calls between Hassanshahi and Sheikhi between August 24 and September 6 (or after October 6), which would be rather remarkable. Patterson’s December affidavit provided no details about the date of the single call discovered using what he identified as DEA’s database, but did specify that the call was made by Hassanshahi’s phone, outbound to Iran. (Patterson didn’t address the later Google production, as that was pursuant to a subpoena.)
To sum up, before Edward Snowden’s leaks alerted us to the scope of NSA’s domestic and international dragnet, Akronowitz claimed he personally had searched multiple databases and found evidence of multiple calls between Hassanshahi’s phone number and Sheikhi’s number, as well as (after getting a month of call records from Google) multiple calls to another Iranian number over unspecified periods of time. After Snowden’s leaks alerted us to the dragnet, after Dianne Feinstein made it clear the NSA can search on Iranian targets in the Section 215 database, which somehow counts as a terrorist purpose, and after Eric Holder decided to shut down just the DEA dragnet, Akronowitz changed his story to claim he had found just one call between Hassanshahi and Shiekhi, and — after a few more months — just one call from another Iranian number to Hassanshahi. Then, two months later, the government claimed that the only database that ever got searched was the DEA one (the one that had already been shut down) which — Patterson told us — was based on records obtained from “United States telecommunications service providers” via a subpoena.
Before I go on, consider that the government currently claims it used just a single phone call of interest — and the absence of any additional calls in a later months’s worth of call records collected that fall — to conduct a warrantless search of a laptop in a state (CA) where such searches require warrants, after having previously claimed there was a potentially more interesting set of call records to base that search on.
Aside from the government’s currently operative claim that it would conduct border searches based on the metadata tied to a single phone call, I find all this interesting for two reasons.
First, the government’s story about how many databases got searched and how many calls got found changed in such a way that the only admission of an unconstitutional search to the judge, in December 2014, involved a database that had allegedly been shut down 15 months earlier.
Maybe they’re telling the truth. Or maybe Akronowitz searched or had searched multiple databases — as he first claimed — and found the multiple calls he originally claimed, but then revised his story to match what could have been found in the DEA database. We don’t know, for example, if the DEA database permits “hops,” but he might have found a more interesting call pattern had he been able to examine hops (for example, it might explain his interest in the other phone number in Iran, which otherwise would reflect no more than an immigrant receiving a call from his home country).
All of this is made more interesting because of my second point: the US side of the call in question was an Internet call, a Google call, not a telephony call. Indeed, at least according to Patterson’s declaration (records of this call weren’t turned over in discovery, as far as I can tell), Hassanshahi placed the call, not Sheikhi.
I have no idea how Google calls get routed, but given that Hassanshahi placed the call, there’s a high likelihood that it didn’t cross a telecom provider’s backbone in this country (and god only knows how DEA or NSA would collect Iranian telephony provider records), which is who Patterson suggests the calls came from (though there’s some room for ambiguity in his use of the term “telecommunications service providers”).
USAT’s story on this dragnet suggests the data all comes from telephone companies.
It allowed agents to link the call records its agents gathered domestically with calling data the DEA and intelligence agencies had acquired outside the USA. (In some cases, officials said the DEA paid employees of foreign telecom firms for copies of call logs and subscriber lists.)
Instead of simply asking phone companies for records about calls made by people suspected of drug crimes, the Justice Department began ordering telephone companies to turn over lists of all phone calls from the USA to countries where the government determined drug traffickers operated, current and former officials said.
Former officials said the operation included records from AT&T and other telecom companies.
But if this call really was placed from a Google number, it’s not clear it would come up under such production, even under production of calls that pass through telephone companies’ backbones. That may reflect — if the claims in this case are remotely honest — that the DEA dragnet, at least, gathered call records not just from telecom companies, but also from Internet companies (remember, too, that DOJ’s Inspector General has suggested DEA had or has more than one dragnet, so it may also have been collecting Internet toll records).
And that — coupled with the government’s evolving claims about how many databases got checked and how many calls that research reflected — may suggest something else. Given that the redactions on the providers obliged under the Section 215 phone dragnet orders haven’t changed going back to 2009, when it was fairly clear there were just 3 providers (AT&T, Sprint, and Verizon), it may be safe to assume that’s still all NSA collects from. A never-ending series of leaks have pointed out that the 215 phone dragnet increasingly has gaps in coverage. And this Google call would be precisely the kind of call we would expect it to miss (indeed, that’s consistent with what Verizon Associate General Counsel — and former DOJ National Security Division and FBI Counsel — Michael Woods testified to before the SSCI last year, strongly suggesting the 215 dragnet missed VOIP). So while FISC has approved use of the “terrorist” Section 215 database for the terrorist group, “Iran,” (meaning NSA might actually have been able to query on Sheikhi), we should expect that this call would not be in that database. Mind you, we should also expect NSA’s EO 12333 dragnet — which permits contact chaining on US persons under SPCMA — to include VOIP calls, even with Iran. But depending on what databases someone consulted, we would expect gaps in precisely the places where the government’s story has changed since it decided it had searched only the now-defunct DEA database.
Finally, note that if the government was sufficiently interested in Sheikhi, it could easily have targeted him under PRISM (he did have a GMail account), which would have made any metadata tied to any of his Google identities broadly shareable within the government (though DHS Inspectors would likely have to go through another agency, quite possibly the CIA). PRISM production should return any Internet phone calls (though there’s nothing in the public record to indicate Sheikhi had an Internet phone number). Indeed, the way the NSA’s larger dragnets work, a search on Sheikhi would chain on all his correlated identifiers, including any communications via another number or Internet identifier, and so would chain on whatever collection they had from his GMail address and any other Google services he used (and the USAT described the DEA dragnet as using similarly automated techniques). In other words, when Akronowitz originally said there had been multiple “telephone calls,” he may have instead meant that Sheikhi and Hassanshahi had communicated, via a variety of different identifiers, multiple times as reflected in his search (and given what we know about DEA’s phone dragnet and my suspicion they also had an Internet dragnet, that might have come up just on the DEA dragnets alone).
The point is that each of these dragnets will have slightly different strengths and weaknesses. Given Akronowitz’ original claims, it sounds like he may have consulted dragnets with slightly better coverage than just the DEA phone dragnet — either including a correlated DEA Internet dragnet or a more extensive NSA one — but the government now claims that it only consulted the DEA dragnet and consequently claims it only found one call, a call it should have almost no reason to have an interest in.
Mieke Eoyang, the Director of Third Way’s National Security Program, has what Ben Wittes bills as a “disruptive” idea: to make US law the exclusive means to conduct all surveillance involving US companies.
But reforming these programs doesn’t address another range of problems—those that relate to allegations of overseas collection from US companies without their cooperation.
Beyond 215 and FAA, media reports have suggested that there have been collection programs that occur outside of the companies’ knowledge. American technology companies have been outraged about media stories of US government intrusions onto their networks overseas, and the spoofing of their web pages or products, all unbeknownst to the companies. These stories suggest that the government is creating and sneaking through a back door to take the data. As one tech employee said to me, “the back door makes a mockery of the front door.”
As a result of these allegations, companies are moving to encrypt their data against their own government; they are limiting their cooperation with NSA; and they are pushing for reform. Negative international reactions to media reports of certain kinds of intelligence collection abroad have resulted in a backlash against American technology companies, spurring data localization requirements, rejection or cancellation of American contracts, and raising the specter of major losses in the cloud computing industry. These allegations could dim one of the few bright spots in the American economic recovery: tech.
How about making the FAA the exclusive means for conducting electronic surveillance when the information being collected is in the custody of an American company? This could clarify that the executive branch could not play authority shell-games and claim that Executive Order 12333 allows it to obtain information on overseas non-US person targets that is in the custody of American companies, unbeknownst to those companies.
As a policy matter, it seems to me that if the information to be acquired is in the custody of an American company, the intelligence community should ask for it, rather than take it without asking. American companies should be entitled to a higher degree of forthrightness from their government than foreign companies, even when they are acting overseas.
Now, I have nothing against this proposal. It seems necessary but wholly inadequate to restoring trust between the government and (some) Internet companies. Indeed, it represents what should have been the practice in any case.
Let me first take a detour and mention a few difficulties with this. First, while I suspect this might be workable for content collection, remember that the government was not just collecting content from Google and Yahoo overseas — they were also using their software to hack people. NSA is going to still want the authority to hack people using weaknesses in such software, such as it exists (and other software companies probably still are amenable to sharing those weaknesses). That points to the necessity to start talking about a legal regime for hacking as much as anything else — one that parallels what is going on with the FBI domestically.
Also, this idea would not cover the metadata collection from telecoms which are domestically covered by Section 215, which will surely increasingly involve cloud data that more closely parallels the data provided by FAA providers but that would be treated as EO 12333 overseas (because thus far metadata is still treated under the Third Party doctrine here). This extends to the Google and Yahoo metadata taken off switches overseas. So, such a solution would be either limited or (if and when courts domestically embrace a mosaic theory approach to data, including for national security applications) temporary, because some of the most revealing data is being handed over willingly by telecoms overseas.
Back in 1993, cartoonist Peter Steiner famously captured a largely held belief about the Internet: “On the Internet, no one knows you’re a dog.”
According to a fascinating new study from Pew, that’s no longer true.
Just 24% of adults “agree” (20%) or “strongly agree” (3%) with the statement: “It is easy for me to be anonymous when I am online.” By contrast, 74% “disagree” (52%) or “strongly disagree” (22%) that it is easy for them to be anonymous.
The poll suggests this is partly because of coverage of government spying, and partly because of corporate spying.
I find two other things about this most interesting. First, the demographics on the specific answers are very fascinating. Just as one example, more affluent people are more likely to check how they come up on Internet searches.
Self-searching activity varies greatly across different groups, particularly by age, income, and household education. Adults under the age of 50 are far more likely to be “self-searchers” than those ages 50 and older, and adults with higher levels of household income and education stand out as especially likely to check up on their own digital footprints.
But I can imagine that’s because they live more of their life online (and they’re more apt to use things like Linked In to apply for jobs). There are also demographic differences in what people find sensitive (see differences in sensitivity about email content at 50, for example). Again, that may reflect the degree to which these tools are available, and therefore are likely to include sensitive communications.
The other thing, however, is that people appear far less worried about metadata than they should be. I get why people are almost universally worried about social security privacy — and this likely reflects the fact that the most immediate threat to everyone is identity theft, not government spying or abuse from Google. But in both government and commercial hands, metadata have become more revealing than content. Respondents don’t seem to worry about it though.