NSA’s Latest Claim: It Only Gets 30% of “Substantially All” the Hay in the Haystack

SIGINT and 215In December 2007, the FBI began intercepting MOALIN’s cell phone.

FBI search warrant affidavit seeking (among other things) additional cell phones, October 29, 2010

Yesterday, Siobhan Gorman reported that NSA’s “phone-data program” collects 20% or less of the phone data in the US. She explains that the program doesn’t collect cell phone data, and so has covered a decreasing percentage of US calls over the last several years.

The National Security Agency’s phone-data program, which has been at the center of controversy over the NSA’s surveillance operations, collects information from about 20% or less of all U.S. calls—much less than previously described by lawmakers.

The program had been described as collecting records on virtually every phone call placed in the U.S., but in fact, it doesn’t cover records for most cellphones, the fastest-growing sector in telephony and an area where the agency has struggled to keep pace, according to several people familiar with the program.

Ellen Nakashima’s report places the percentage between 20 and 30%, echoing Gorman’s claim about limits on cell data.

The actual percentage of records gathered is somewhere between 20 and 30 percent and reflects Americans’ increasing turn away from the use of land lines to cellphones. Officials also have faced technical challenges in preparing the NSA database to handle large amounts of new records without taking in data such as cell tower locations that are not authorized for collection.


The bulk collection began largely as a land-line program, focusing on carriers such as AT&T and Verizon Business Network Services. At least two large wireless companies are not covered — Verizon Wireless and T-Mobile U.S., which was first reported by the Wall Street Journal.

Industry officials have speculated that partial foreign ownership has made the NSA reluctant to issue orders to those carriers. But U.S. officials said that was not a reason.

“They’re doing business in the United States; they’re required to comply with U.S. law,” said one senior U.S. official. “A court order is a court order.”

Rather, the official said, the drop in collection stems from several factors.

Apart from the decline in land-line use, the agency has struggled to prepare its database to handle vast amounts of cellphone data, current and former officials say. For instance, cellphone records may contain geolocation data, which the NSA is not permitted to receive.

These reports offer a more credible explanation than Geoffrey Stone’s multiple claims to this effect about why the program misses data. So they may be true.

But I think they instead point to the legal range of authorities NSA uses to collect phone records, not to what records they actually have in their possession.

These reports are commenting (though without specifying, or even seeming to be aware they need to specify) on what the government claims it collects under Section 215. These reports are not commenting on what NSA collects under all authorities.

In this post I will show why I believe these reports to be credible only in a very narrow sense. In a follow-up post I will point to the legal issues that underlie the Administration’s conflicting claims about what it collects.

Read more

Is NSA Wiretapping Now Rather than Tipping?

One of the news bits a number of outlets took away from the phone dragnet order document dump 10 days ago is that the NSA averages(d) about 3 tips a day to the FBI.

That’s actually not news. It’s consistent with a series of accountings NSA gave to Reggie Walton in 2009, as when, in February 2009, they provided more exact numbers (though they’d get tweaked a bit during that summer) that were smaller, but still in the range of 2-3 tips a day.

Demonstrating the value of the BR metadata to the U.S. Intelligence Community, the NSA has disseminated 275 reports and tipped over 2,500 telephone identifiers to the FBI and CIA for further investigative action since the inception of this collection in docket number BR 06-05.

That said, at least according to Geoffrey Stone, the scale of the referrals may have gone down dramatically.

Under the FISA statute, the NSA queried 288 numbers in 2012 and had only 16 instances where matches were analyzed, confirmed, and then forwarded to the FBI. According to Stone, these queries only produced about 6,000 numbers that were “touched” by the analysis, of the millions of numbers whose meta-data the NSA stores for up to five years.

In general and specifically here, there are reasons I don’t entirely trust Stone’s comments on the dragnet. He has said a lot that is inconsistent with other public (and legally sworn) claims, notably on the volume of phone records collected. And his silences about certain aspects of the dragnet make me wonder how complete an understanding he has.

Plus, the “16 instances” may — as was true in the earlier period — represent reports that include more than one number. If, as occurred until 2009, each report had roughly 10 numbers, then this might amount to 160 identifiers (which is still far below the pace of the 2006-2009 period, but then during that period they weren’t enforcing RAS).

Then there’s the complete lack of definition for “touch” with regards to his 6,000 number.

In addition, 2012 might be a new baseline (or perhaps outlier) year, as the rollout of the new automated system at the end of 2011 would likely have changed the treatment of phone identifiers entirely.

And as I’ve said, I expect the use of the phone dragnet for a “peace of mind” query after the Boston Marathon attack to result in a huge number of tips (though perhaps in just one or several reports), given how wired the Tsarnaevs were and had been for the five years leading up to the attack.

Moreover, in a development that may or may not be entirely unrelated, the number of telephone taskings under Section 702 have started to go up again starting in 2012, after having been down since 2009.

As the chart demonstrates, the number of newly tasked telephone numbers decreased after 2009, but began to increase again in 2012. The average number of telephone numbers tasked each month for the first 11 months of 2012 [redacted].

There are admittedly a number of possible explanations (increasing collection of text messages, different kind of upstream collection, potentially even a fourth certificate in addition to the terror, proliferation, and cyber ones we know about). But one possibility is that the new alert system has led NSA to move toward wiretapping interesting numbers, rather than sending them to FBI for investigation. Moreover, by wiretapping someone, NSA could share data with FBI and CIA in relatively unfettered fashion, as both are permitted to receive unminimized content under 702 in certain circumstances, and both have the authority to do backdoor searches on US person content on all but upstream collected 702 data.

The NSA can’t give phone numbers to FBI without review, but according to section 702 minimization procedures, in some cases they can let CIA and FBI read wiretap content without such review.

That is, wiretapping someone could be a way to evade data dissemination restrictions in place on actual phone dragnet queries.

If by “Big Data” You Mean “Big Campaign Donations”

President Obama has named the people who will help John Podesta accomplish this task.

I have also asked my Counselor, John Podesta, to lead a comprehensive review of big data and privacy. This group will consist of government officials who—along with the President’s Council of Advisors on Science and Technology—will reach out to privacy experts, technologists and business leaders, and look at how the challenges inherent in big data are being confronted by both the public and private sectors; whether we can forge international norms on how to manage this data; and how we can continue to promote the free flow of information in ways that are consistent with both privacy and security.

As I said in my annotations to Obama’s speech, effectively Obama responded “to a review by calling for another review,” but at least it would be a welcome first time he reached out to technologists.

Here’s the list:

That’s why in his speech, the President asked me to lead a comprehensive review of the way that “big data” will affect the way we live and work; the relationship between government and citizens; and how public and private sectors can spur innovation and maximize the opportunities and free flow of this information while minimizing the risks to privacy. I will be joined in this effort by Secretary of Commerce Penny Pritzker, Secretary of Energy Ernie Moniz, the President’s Science Advisor John Holdren, the President’s Economic Advisor Gene Sperling and other senior government officials.

I’ll outsource judging whether this amounts to reaching out to technologists to Chris Soghoian:

None of the big names named in the president’s “big data” review announcement are technologists. DC at its finest.

But I’m particularly interested in Penny Pritzker’s presence on the list.

After Cass Sunstein and Geoffrey Stone ended up being too independent to deliver the whitewash Obama wanted, he has picked one of his biggest campaign donors to review Big Data.

So I guess by “Big Data” we know what Obama meant.

Worse still, Pritzker heads up an Agency that — it is increasingly clear — serves a key role in offering carrots and sticks to coerce compliance from private companies with government data demands. And compliance not just for the purposes of defense of spying, but also for cyberoffense. Not exactly the kind of person who might expect candor from the Big Data companies likely to be coerced by the government.

The Government Plays Connect-the-Dots Differently than They Say

In my continuing obsession to understand precisely how the government really uses the dragnet, consider this post, in which NSA Review Group member Geoffrey Stone conducts (IMO) inadequate analysis to conclude the phone dragnet is probably unconstitutional.

In it, he provides this description of how the government uses the phone dragnet:

In 2012, the NSA queried a total of 288 phone numbers. Based on these queries, the NSA found 16 instances in which a suspect phone number was directly or indirectly in touch with another phone number that the NSA independently suspected of being associated with terrorist activity. In such cases, the NSA turns the information over to the FBI for further investigation.

In terms of the “connect the dots” metaphor, the purpose of the program is not so much to discover new “dots” but to determine if there are connections between two or more already suspect “dots.” For example, if a phone number belonging to a terrorist suspect in Pakistan is found to have called a phone number in the United States that the government independently suspects belongs to a person involved in possible terrorist activity, alarm bells (figuratively) go off very loudly, alerting the government to the need for immediate attention. [my emphasis]

I don’t think this can be an accurate description of how the dragnet works.

It is close to what happened with Adis Medunjanin. As the FBI was honing in on Najibullah Zazi, the NSA did a query and found a new cell phone for Medunjanin, though they already knew Medunjanin was a likely accomplice of Zazi’s through via travel records. The government says they were particularly interested in this phone because it was in contact with other extremists. Thus, they found a brand new phone number, but one that ended up being associated with both a suspect (Medunjanin) and other suspects (the other people that phone was in contact with).

But that cell phone for Medunajnin was a brand new number to the NSA, at least according to their reports.

The claim may still be true if they used burner matching to identify Medunjanin as a match to the other phone record they had on him. But it seems this process would have to involve additional information about Medunjanin at some point — at the very least, the match of those travel documents to that phone number, if not his identity.

In other words, this only seems to make sense if they had Medunjanin’s “identity” in some form or another, belying their claims not to have identities while they’re contact chaining.

The description is potentially more problematic with Basaaly Moalin. In his case, the stated explanation for what happened is they found his number on a second-degree search, sent it to the FBI, and the FBI learned he was the guy who had previously been investigated in 2003.

The problem might be alleviated in two ways: first, if the hawala through which Moalin was sending money to Ayro, was also tied to a suspect number. That’s a distinct possibility: but the question is, how does that identity as a suspect number get communicated to NSA? If NSA already had it, doesn’t it mean they’ve got more suspect numbers sitting somewhere than have been RAS approved?

The other possibility is that Moalin himself was still identified as a suspect number from the investigation back in 2003 — that an investigation that turned up no evidence might still, during the era of the illegal program, have gotten someone nominated as a suspect number under Cheney’s program, and they never purged the system entirely (which would seem to be supported by the 2009 problems, which showed they hadn’t turned off the illegal program features).

Either of these possibilities, of course, would raise new concerns about the NSA program.

But the description would also raise real issues, both about the honesty of witnesses and the potential efficacy of the system. If the NSA only triggers on people who’ve got ties to a second suspect number (which is entirely different than what they’ve been saying) then it could not possibly alert the government to a fully compartmented lone actor (someone like, say, Faisal Shahzad). That is, it would only find people who were engaged in the kind of elaborate planning seen before the government dismantled al Qaeda, but would not find the kind of individual extremists we’ve seen almost exclusively (with the exception of Zazi) for years.

This would answer the question of whether the NSA is finding the right numbers, in that it would be less likely to find someone innocent. It also might explain why the program didn’t find Shahzad. But it would also mean it does (as presented) far less than the NSA has been saying it does.

I don’t actually believe that, but that is what it would suggest.

Sucky Assessments of the Phone Dragnet Reveal How Much They’re Keeping “Secret”

The assessments of the phone dragnet suck.

I don’t mean the assessments of the phone dragnet show the program sucks, though that may well be the case. I mean the assessments of the phone dragnet I’ve seen do a very poor job of assessing the value of it. Which serves to show how much of the larger dragnet remains, if not secret, still largely undiscussed.

To see what I mean, consider this post, from Just Security’s Ryan Goodman.

Insiders disagree about the phone dragnet value with outsiders

The strongest part of his post compares the seemingly contradictory assessments of the phone dragnet by two different members of the NSA Review Group. University of Chicago Professor Geoffrey Stone and Deputy Director of CIA Mike Morell.

Stone, based on what he learned from public sources and from the briefings the Group received, believes the program did not prevent any terrorist attacks. Morell, whose former agency receives Tippers from the program and even had direct access to query results until 2009 just like the FBI does and did (though no one talks about that) insists it has helped prevent terrorist attacks.

Goodman also notes that the Gang of Four immediately defended the phone dragnet after the Review Group released its results (actually, they object to more than the phone dragnet recommendation but don’t say what other recommendations they object to), but doesn’t note the terms they use to do so:

However, a number of recommendations in the report should not be adopted by Congress, starting with those based on the misleading conclusion that the NSA’s metadata program is ‘not essential to preventing attacks.’ Intelligence programs do not operate in isolation and terrorist attacks are not disrupted by the work of any one person or program. The NSA’s metadata program is a valuable analytical tool that assists intelligence personnel in their efforts to efficiently ‘connect the dots’ on emerging or current terrorist threats directed against Americans in the United States. The necessity of this program cannot be measured merely by the number of terrorist attacks disrupted, but must also take into account the extent to which it contributes to the overall efforts of intelligence professionals to quickly respond to, and prevent, rapidly emerging terrorist threats. [my emphasis]

In other words, Goodman presents evidence that the Gang of Four and a former top CIA official believe there are other reasons the phone dragnet is valuable, while someone relying on limited briefings evaluates the program based on its failure to stop any attack.

That ought to make Goodman ask what Morell and Dianne Feinstein know (or think they know) that Stone does not. It ought to make him engage seriously with their claim that the phone dragnet is doing something else beyond providing the single clues to prevent terrorist attacks.

One they’re not willing to talk about explicitly.

Assessments and the terrorist attack thwarted metric

Instead, Goodman assesses the phone dragnet solely on the basis of the public excuse offered over and over and over since the Guardian first published the Verizon order in June: to see which Americans are in contact with (alleged) terrorist associates so as to prevent an attack.

Goodman lectures program critics that identifying funders or members of terrorist groups might help find terrorists, too, and “peace of mind” might help dedicate resources most productively.

The key objective of course is to stop terrorist attacks against the US homeland and vital US interests abroad. An important distinction, however, is whether the intelligence generated by the program is:

(a) “direct”: timely information to foil a specific attack; or

(b) “indirect”: information that enables the government to degrade a terrorist group or decrease the general likelihood of attacks

Examples of the latter might include information on individuals who have joined or are funding a terrorist organization. Intelligence could help to identify and successfully prosecute such individuals, and hence disable them and deter others. The important point is that both types of information aid the overall goal of stopping terrorist attacks. That point appears to have been lost on some critics of the program. When the government cites the latter information yields, critics often consider such situations irrelevant or little to do with stopping attacks.

But Goodman imagines only those affirmatively supporting terrorism would help the government prevent terrorism, which is not necessarily the case.

Does the NSA’s network analysis even pick the right calls?

One thing missing from such assessments are the failures. Why didn’t, for example, Faisal Shahzad’s planning with the Pakistani Taliban identify him and his hawala before the attack? There are plausible explanations: he used good enough operational security such that he had no communications that could have included in the dragnets, his TTP phone and Internet contacts were not among the services sucked up, the turmoil in the phone and (especially) Internet dragnet in 2009 and 2010 led to gaps in the collection. Then there’s a far more serious one: that the methods NSA use to identify numbers of interest may not work, and may instead only be identifying those whose doings with terror affiliates are relatively innocent, meaning they don’t use operational security (though note the US-based phone dragnets would use more sophisticated analysis only after data gets put in the corporate store, whereas data collected overseas might be immediately subject to it).

And for those who, like Goodman, place great stock in the dragnet’s “peace of mind” metric, they need to assess not just the privacy invasion that might result, but the resources required to investigate all possible leads — which could have been upwards of 36,000 people in the Boston Marathon case.

That is, unless we have evidence that NSA’s means of picking the interesting phone contacts from the uninteresting ones works (and given the numbers involved, we probably don’t have that), then the dragnet may be as much a time suck as it is a key tool.

What about the other purposes the Intelligence Community has (quietly) admitted?

The other problem with assessments of the phone dragnet is they don’t even take the IC at its word in its other, quieter admissions of how it uses the dragnet (notably, in none of Stone’s five posts on the dragnet does he mention any of these — one, two, three, four, five — raising questions whether he ever learned or considered them). These uses include:

  • Corporate store
  • “Data integrity” analysis
  • Informants
  • Index

Corporate store: As the minimization procedures and a few FISC documents make clear, once the NSA has run a query, the results of that query are placed in a “corporate store,” a database of all previous query results. Read more

The No-Technologist Technology Review Panel

In addition to the four people ABC earlier reported would be part of Obama’s Committee to Learn to Trust the Dragnet, Obama added … another law professor, Geoffrey Stone. (Stone is [see update], along with Swire, a worthwhile member. But not a technologist.)

What’s fucking crazy about the committee is it has zero technologists to review a topic that is highly technical. Obama implicitly admits as much! He sells this committee for their “immense experience in national security, intelligence, oversight, privacy and civil liberties.” National security, intelligence, oversight, privacy, civil liberties. No technology.

On August 9, President Obama called for a high-level group of experts to review our intelligence and communications technologies. Today the President met with the members of this group: Richard Clarke, Michael Morell, Geoffrey Stone, Cass Sunstein and Peter Swire.

These individuals bring to the task immense experience in national security, intelligence, oversight, privacy and civil liberties. The Review Group will bring a range of experience and perspectives to bear to advise the President on how, in light of advancements in technology, the United States can employ its technical collection capabilities in a way that optimally protects our national security and advances our foreign policy while respecting our commitment to privacy and civil liberties, recognizing our need to maintain the public trust, and reducing the risk of unauthorized disclosure.

The President thanked the Members of the Group for taking on this important task and looks forward to hearing from them as their work proceeds. Within 60 days of beginning their work, the Review Group will brief their interim findings to the President through the Director of National Intelligence, and the Review Group will provide a final report and recommendations to the President. [my emphasis]

So in spite of the fact that the White House highlights technology in its mandate, that didn’t lead them to find even a single technologist.

Also: Cass Sunstein.

Also: the Committee does, in fact, report its findings through James Clapper, the guy whose programs they will review, they guy who lied to Congress.

At least the White House isn’t promising — as Obama originally did — that it will be an “outside” “independent” committee.

Update: Egads. I take back what I said about Stone, who said this in June.

[W]hat should Edward Snowden have done? Probably, he should have presented his concerns to senior, responsible members of Congress. But the one thing he most certainly should not have done is to decide on the basis of his own ill-informed, arrogant and amateurish judgment that he knows better than everyone else in government how best to serve the national interest. The rule of law matters, and no one gave Edward Snowden the authority to make that decision for the nation. His conduct was more than unacceptable; it was criminal.