What the End of Bulk Metadata Collection Would Mean for Intelligence Collection

This story has been updated to reflect late-breaking developments. 

The future of the NSA’s bulk metadata collection program is in serious doubt, which raises the questions: how useful is it to the intelligence community, and what will they do if it goes away? 

Less than two years after the first Snowden revelations, President Obama this week endorsed a bill to end the government’s practice of sucking up and storing one kind of electronic data — information about Americans’ telephone calls. The bill, called the Freedom Act, passed the House on Wednesday, May 13, by a vote of 338 to 88. Meanwhile senators from both parties threatened to filibuster the renewal of the Patriot Act, which provides for the collection of bulk telephone metadata. And one federal court has ruled the practice unlawful. 

Update: the Senate failed to reauthorize key provisions of the Patriot Act by a June 1st deadline, temporarily throwing various intelligence collection activities into a state of limbo.  White House Press Secretary, Josh Earnest, responded: "We call on the Senate to ensure this irresponsible lapse in authorities is as short-lived as possible." The Senate passed the Freedom Act on Tuesday.

What would ending bulk collection mean for intelligence collection? Last year, the National Academy of Sciences, or NAS, began looking at what would happen if U.S. spies no longer had easy access to phone records. In January, they issued their findings as Bulk Collection of Signals Intelligence: Technical Options. The 80-page report draws open a curtain on the probable future of electronic spying.

The most important conclusion comes early on: a huge database of the phone records of millions of people is quite valuable to signals intelligence, or SIGINT. “There is no software technique that will fully substitute for bulk collection where it is relied on to answer queries about the past after new targets become known,” the NAS report says.

But the report also suggests that there may be ways to at least partially replace this. “It may be possible to improve targeted collection [as opposed to bulk collection] to the point where it provides a viable substitute for bulk collection in at least some cases, using profiles of potential targets that are compiled from a wide range of information.”

What does that mean? Metadata can be defined, broadly, as all the data that you produce through your exchanges with digital devices. On its most literal level, it’s “a set of data that describes and gives information about other data.” For example, metadata about a phone conversation might include the numbers of the caller and the recipient, where each one was when they began the call, and who they called previously or next—but not an actual recording of the conversation.

The NSA sweeps up up vast amounts of these digital identifiers, including information about the activities of millions of Americans not related to any particular investigation or target, and holds them for up to five years. It does this under the authority of the Foreign Intelligence Surveillance Act (FISA)’s Section 215, which was enacted as part of the USA Patriot Act in 2001.

What does the intelligence community get from all the metadata it collects and stores? The report outlines three broad areas, starting with “contact chaining.” Phone numbers, time logs, and other pieces of metadata can be used to find hidden connections between people: middlemen, hidden contacts, or simply mutual acquaintances. Next is alternate identifier discovery, aka finding digital aliases: names, usernames, and even different communications methods a person uses. Lastly, there is triage, or ranking the urgency of threats.

The larger the trove of metadata on hand, and the farther an individual’s “stream” extends backward into time, the more opportunity there is to find potentially useful information. The amount of bulk metadata held by the NSA — telephone and otherwise — is classified. But a 2007 report from the Defense Department, first reported by James Bamford in Wired, suggested that DoD is building enough storage to keep one yottabyte’s worth of data, or one septillion bytes.

On Tuesday, May 12, the White House endorsed passage of the USA Freedom Act, which the House passed on Wednesday. If it ultimately becomes law, which is looking very likely, it would shift bulk-data-collection duties to the phone companies, which would provide information to the intelligence community upon presentation of secret warrants. Some have argued that bulk collection could continue under the Freedom Act, depending on how warrant requests are interpreted by FISA court judges, who almost always approve the spies’ requests. Regardless, the White House, and Freedom Act proponents saw their case bolstered when a U.S. Appeals Court declared unlawful the practice of scooping up telephone records of millions of Americans. Two other appeals courts are also looking at the issue, which could ultimately wind up before the Supreme Court.

Spotify for Spies: Replacing A Valuable Tool

Concerns about privacy and civil society aside, the NAS report says (passive aggressively) ending bulk collection would eliminate a unique and valuable tool.

“If the past events are unique or if delay in obtaining results is unacceptable (because of an imminent threat or perhaps because of press coverage or public demand), then the intelligence will not be as complete. So restricting bulk collection will make intelligence less effective, and technology cannot do anything about this; whether the gain in privacy is worth the loss of information is a policy question that the committee does not address,” the authors write.

But the report also explores a partial remedy; a conceptual system that would give analysts access to stored metadata only under certain circumstances and under tightly controlled limits. Think of a streaming music service like Pandora or Spotify that allows you to listen to a song but prevents you from downloading it. In the consumer world, this is called “digital rights management”; in the context of intelligence collection, it’s “usage control.” No matter where the phone-record databases live — at the phone company or deep within the NSA — usage control seems likely to become part of intelligence collection.

This raises another question: how should analysts figure out what portions of the data to ask for? It’s hard to ask for the “right” puzzle piece if you don’t know what they all look like.

Again, the report offers a partial solution: create a new generation of artificial intelligence agents to assess the relevance of data either in real time or in storage. “More powerful automation could improve the precision, robustness, efficiency, and transparency of the controls, while also reducing the burden of controls on analysts,” the report says.

Such a system would, in theory, prevent unlawful fishing expeditions by limiting the scope, queries, and even personnel able to access to the data. “The goal is to do this well enough that software can decide which queries are allowed by the policy, or at least drastically reduce the number of queries that require manual, human approval. This is certainly feasible for limited classes of queries such as ‘Find all the phone numbers that have connected in the last month to this list of numbers belonging to a known target.’ Indeed NSA already has pre-approved queries,” says the report.

Finally, the program would keep a log of all the above so that overseers could make sure that analysts weren’t abusing the system … or absconding with the royal jewels to Hong Kong. “In each of these areas, there are opportunities for automated control; some of them are already deployed in the [intelligence community] or in private companies, some have been demonstrated in research laboratories, and some are promising research directions,” says the report.

The report highlights the Security and Privacy Assurance Research (SPAR) program at the Intelligence Advanced Research Projects Agency, or IARPA. Begun in 2011, SPAR developed cryptographic tools to restrict database queries. Such a system might scour digital records belonging to businesses, such as time and call logs maintained by phone companies on their customers, and then show those records to intelligence analysts under specific conditions.

“Today, NSA says it cannot collect any sizeable fraction of all global communications data, and it may likewise be that despite declining computing costs, NSA will not be able to automatically analyze more than a tiny bit. However, in many cases, the operators of the sensors will apply the algorithms to meet business needs, such as identifying license plates to bill parking charges. In these cases, the analyzed data may be available to NSA in the form of business records,” says the report.

Google vs. NSA?

Technically, this problem isn’t substantially different than other AI challenges that have recently occupied the best minds of Silicon Valley. Machines have learned to find YouTube videos featuring cats and to recognize how some queries may indicate a user’s interest in a particular product. Both are Google projects; the latter allows the search giant to send you targeted ads in Gmail without identifying you to advertisers.

Americans (generally) trust Gmail with their data because they know that humans at Google headquarters aren’t reading their emails. Will privacy advocates be okay with letting smarter algorithms scan business records (telephone metadata held by companies) for data with “intelligence value” and then deciding what to pass along to humans? The authors of the report certainly hope so: “Perhaps if the public comes to embrace the philosophy and practice of usage controls for sensitive personal data, such as health and financial data, and comes to trust private sector IT implementations of the protection procedures, controlled-use approaches to intelligence information can find greater favor.”

Yet this begs another question: when does an algorithm become so smart that it constitutes a human-like threat to privacy? It depends on the degree to which the program understands the data that it’s collecting. A 2013 article in The Futurist magazine outlines how a supercomputer, under a Defense Advanced Research Project Agency grant, was taught to comprehend the meaning of sentences based on grammar and syntax. It performed at 85% accuracy

The article contains a response by Peter Eckersley — today, the chief computer scientist at the Electronic Frontier Foundation, a privacy watchdog group — that anticipates the debate to come. “It’s intrusive to have algorithms scanning your email. And it’s more intrusive to have smarter algorithms scanning your email. We should be concerned about the privacy implications of both of these cases, but the danger clearly gets worse as the software gets smarter.”

What happens to your phone metadata, whether it stays with your phone companies or goes to the NSA data storage center in Utah, is a matter of temporary concern. The future of intelligence collection belongs to the machines. That’s where the debate is headed as well.