The Surveillance Architecture

The Infrastructure Before the Products

The standard account of Silicon Valley’s emergence treats it as a story of civilian entrepreneurship — venture capital, garage startups, brilliant engineers solving problems the market had not yet defined. The documented record runs differently. The infrastructure on which the consumer internet was built — packet-switching networks, anonymizing relay protocols, satellite geospatial imaging, data-integration platforms for behavioral analysis — was developed inside or directly adjacent to the U.S. defense and intelligence apparatus, and the commercial companies that became dominant in each space entered the market either through intelligence-community seed investment, through the acquisition of government-funded research, or through the retroactive integration of their platforms into classified surveillance programs that their leadership publicly denied awareness of.

The architecture is not a secret in the sense of concealment through opacity. Its major components — the CIA’s venture capital arm, the NSA’s collection programs, the Five Eyes intelligence-sharing arrangement, the DARPA research pipeline — are documented in primary sources: government press releases, declassified agreements, FISA court orders published under legal challenge, and 1.7 million documents provided to journalists by NSA contractor Edward Snowden beginning in June 2013. What the public record requires is not further disclosure so much as sequence. Read in isolation, each component carries its own institutional rationale, its own legal authorization, and its own market logic. Read in sequence, they form a closed loop in which intelligence agencies fund foundational technology, acquire access to the companies that commercialize it, compel further cooperation through legal architecture, circumvent remaining legal constraints through allied-nation intelligence sharing, and integrate the resulting data streams into searchable, actionable analytical platforms — all while the nominally private companies present themselves to the public as independent actors whose data practices are, occasionally, subject to legal misuse by government actors they would prefer to resist.

The strongest objection to this reading is that it attributes coordination where convergent institutional incentive is sufficient. The response is that convergent institutional incentive producing coordinated outcomes across three administrations, multiple agencies, and several allied governments is the functional definition of coordination, regardless of whether any conspiracy meeting was ever minuted.

In-Q-Tel and the CIA’s Technology Acquisition Strategy

On September 29, 1999, a 501(c)(3) nonprofit corporation was incorporated in Virginia under the temporary name “Peleus,” subsequently renamed In-Q-It and then In-Q-Tel. Its founding was commissioned by CIA Director George Tenet, whose agency had engaged Business Executives for National Security (BENS) the prior year to assess the intelligence community’s technological capacity relative to the commercial sector. The BENS assessment concluded the CIA was running fifteen to twenty years behind commercial technology development — a structural deficit that conventional government procurement, with its classified specifications and multi-year contracting cycles, could not close. The structural solution was to operate as a venture capital firm rather than a procurement office.

The mechanism contains several compounding elements that distinguish it from ordinary investment. The CIA funds In-Q-Tel through budget line items that are not separately disclosed in public appropriations. In-Q-Tel invests in early-stage technology companies, takes equity positions, and functions as a conventional venture capital actor — board seats, follow-on rounds, market-rate returns sought. The investment, however, comes paired with a Technology Access Agreement: a contractual requirement that the portfolio company make its product available to the CIA and the broader intelligence community on specified terms. The investment’s visible face is commercial; its operative content is procurement, structured to appear as market participation. The 501(c)(3) status creates an oversight gap precisely calibrated to this ambiguity: too public to be classified, too opaque to face standard government procurement scrutiny.

A second compounding element is credentialing. An In-Q-Tel investment functions as a signal to government contracting officers and defense primes that a company has been vetted for classified environments. Companies In-Q-Tel touches gain access to procurement channels that ordinary competitive processes cannot open. The initial investment generates a relationship that structures the company’s entire federal market trajectory.

The clearest illustration of how this mechanism operates is the Keyhole Corporation trajectory. Keyhole produced EarthViewer, a commercial geospatial visualization platform that rendered satellite imagery as a navigable three-dimensional surface. In-Q-Tel announced its investment in Keyhole on June 25, 2003, and the National Geospatial-Intelligence Agency adapted the software for government use; television coverage and military users also employed EarthViewer during the Iraq war. Google acquired Keyhole on October 28, 2004 and relaunched the product in June 2005 as Google Earth. The exact sequence matters: intelligence investment and operational adaptation entered a commercial product before a platform acquisition carried the same visual grammar to a mass public. Nothing had to be “laundered” from a nonexistent classified origin. The acquisition simply concealed how close the military, intelligence, venture, and consumer versions had already become.

Geofeedia, another In-Q-Tel portfolio company, provided geolocation-based social media surveillance — monitoring platform activity by physical location to track who was attending which public events and what they were communicating in real time. Law enforcement in Baltimore, Oakland, and Chicago used the platform to monitor Black Lives Matter demonstrations, a function disclosed through ACLU investigation. Facebook, Twitter, and Instagram revoked Geofeedia’s data access after the ACLU made the surveillance use public. The company dissolved in 2017. The episode illustrates the architecture’s extension into domestic political monitoring: In-Q-Tel’s portfolio reached from foreign intelligence collection into the surveillance of American citizens exercising first amendment rights, with the intelligence community’s investment credentials providing the commercial credibility that secured the social platform data access in the first place.

By 2016, In-Q-Tel had invested in over three hundred companies across machine learning, computer vision, natural language processing, biometrics, geospatial analysis, satellite imagery, and network analysis. The portfolio maps the cognitive science of surveillance — every technical domain required to build a system that identifies, locates, and models the behavioral and associational patterns of human populations.

Palantir: Co-Development, Not Investment

The most fully documented case of intelligence-community co-development — rather than mere investment — involves Palantir Technologies. Palantir was founded in 2003 by Peter Thiel, Alex Karp, and Joe Lonsdale, with an initial $2 million investment from In-Q-Tel after mainstream venture firms including Sequoia Capital declined to fund it. The company built data-integration and analytical visualization platforms that could ingest both structured and unstructured data — spreadsheets, communications intercepts, satellite imagery, financial records — and render relationships within that data as manipulable network graphs. The product’s name for the government market was Palantir Gotham.

Two documents published by The Intercept on February 22, 2017, drawing on the Snowden archive, establish the intelligence community’s relationship to Palantir as constitutively different from ordinary investor-portfolio company dynamics. A 2008 GCHQ VisWeek Conference report described Palantir as “a relatively new Silicon Valley startup who are sponsored by the CIA.” A 2011 NSA presentation disclosed that Palantir was “created through iterative collaboration between Palantir computer scientists and analysts from various intelligence agencies over the course of nearly three years.” The intelligence community did not invest in Palantir and then deploy it. Intelligence analysts co-built the product alongside Palantir engineers, shaping its analytical architecture from within, over a multi-year development cycle that preceded the company’s wider commercial release.

The operational integration ran deeper still. GCHQ built a Palantir plugin designated TWO FACE — documented in the agency’s own internal wiki — that allowed analysts to import data from NSA’s XKeyscore surveillance system directly into Palantir for visualization and network analysis. A GCHQ analyst attested in the TWO FACE documentation that using the plugin “reduced the initial analysis time by at least 50%” compared to working in XKeyscore directly. A second plugin called Kite enabled GCHQ to ingest data from “third parties to which GCHQ was able to gain access” into Palantir repositories. By 2011, Palantir was simultaneously connected to at least ten classified Five Eyes and GCHQ programs, embedded into the LOVELY HORSE open-source intelligence processing operation, and being tested for deployment on Android handsets as part of the SIGINT augmented-reality component of the joint Mastering the Internet program.

The commercial framing of Palantir as a Silicon Valley technology company — a company that went public on the NYSE in September 2020 with a market capitalization exceeding $15 billion — is accurate in the narrow sense that it is incorporated, issues equity, and sells products. The intelligence community’s own documents describe something more precisely: a product whose architecture was shaped by the operational needs of classified surveillance programs, deployed within those programs through dedicated integration plugins, and now contracted to domestic law enforcement through Immigration and Customs Enforcement (ICE) contracts whose value exceeded $49 million by 2019, enabling ICE agents to query biographical records, travel history, financial data, and social network analysis through the HSI Investigative Case Management system built on Palantir Gotham.

The DARPA Pipeline

The integration of defense funding into foundational internet infrastructure predates In-Q-Tel by three decades. The ARPANET, the packet-switching network that became an architectural ancestor of the modern internet, was funded by ARPA’s Information Processing Techniques Office and demonstrated operationally on October 29, 1969, when the first message passed between UCLA and Stanford Research Institute. Its immediate purpose was resource sharing among research computers, not a command network designed expressly to survive nuclear war. Survivable distributed communication belonged to the surrounding packet-switching problem, and the civilian internet still inherited decisive protocols, institutions, and engineers from a defense-funded network. The origin is powerful enough without the nuclear-war legend.

Tor — The Onion Router — followed the same trajectory from military application to civilian infrastructure. The core research was conducted by Michael G. Reed, Paul F. Syverson, and David M. Goldschlag at the U.S. Naval Research Laboratory, published in a 1997 IEEE Symposium on Security and Privacy paper and developed through DARPA funding in collaboration with MIT’s Free Haven Project. A 2004 paper by Roger Dingledine, Nick Mathewson, and Syverson formalized the second-generation protocol. The Tor Project incorporated as a nonprofit in 2006 and became the primary anonymity tool used by journalists, dissidents, and human rights workers in authoritarian states. The same NSA, through XKeyscore, later built specific targeting capabilities directed at identifying Tor users — documented in Snowden slides published by The Guardian in July 2014. The Naval Research Laboratory invented the primary tool for evading NSA surveillance. The NSA then built specific capabilities to defeat it. The same funding source underwrote both sides of the evasion-detection dialectic.

The Total Information Awareness program, announced by retired Admiral John Poindexter at a DARPA conference on August 2, 2002, represented the most explicit public statement of the integration’s intended scope. Poindexter — previously convicted on five felony counts in the Iran-Contra affair, with convictions later overturned on appeal — described TIA as “a Manhattan Project for counter-terrorism,” aimed at integrating all available data streams — financial transactions, travel records, communications intercepts, biometric data, medical records — into a unified analytical environment capable of detecting terrorist planning before it produced attacks. Congress defunded the Information Awareness Office through the Consolidated Appropriations Resolution of 2003. The programs themselves migrated. Topsail and Basketball — renamed successors to TIA components including Project Genoa II — were transferred to NSA. The analytical architecture that TIA had articulated became the conceptual blueprint for PRISM and XKeyscore’s operational design within the decade.

The most precisely documented coincidence in the DARPA pipeline involves LifeLog. LifeLog was a DARPA program to build a permanent, searchable record of everything an individual does — every purchase, email, website visited, relationship formed — a total-capture autobiographical database. DARPA cancelled LifeLog on February 4, 2004, citing privacy concerns that had generated sustained opposition from civil liberties organizations. Facebook was registered as a business and launched to Harvard students on February 4, 2004. The same-day coincidence is confirmed in contemporaneous reporting — Noah Shachtman’s Wired piece on the cancellation was published February 4, 2004; Facebook’s launch date is independently documented. The documentary record establishes the coincidence and the functional parallel between LifeLog’s described objectives and Facebook’s realized architecture: a platform that captures communication, relationship, location, behavioral, and preference data for every user and maintains it in searchable, queryable form. LifeLog supplies the design precedent; Facebook supplies the deployed social infrastructure.

The Snowden Disclosures: Program by Program

Between June 5, 2013 and subsequent months of publication, journalists at The Guardian, The Washington Post, and The Intercept published primary documentary evidence from materials provided by NSA contractor Edward Snowden. The documents established, for the first time in confirmed primary-source form, the operational architecture of mass surveillance running across commercial platforms, fiber optic infrastructure, and allied-nation collection systems simultaneously.

The first disclosure, published June 5, 2013, was a Foreign Intelligence Surveillance Court order — Docket BR 13-80, dated April 25, 2013, signed by Judge Roger Vinson — directing Verizon Business Network Services to produce all telephony metadata for calls within the United States and between the United States and abroad, on an ongoing daily basis. The order established that bulk domestic metadata collection was not a theoretical possibility or a privacy advocate’s hypothetical. It was a court-authorized, continuously operating program.

PRISM, disclosed simultaneously by The Guardian and The Washington Post on June 6, 2013, was documented in a forty-one-slide presentation classified TOP SECRET//SI//ORCON//NOFORN, designated US-984XN. Under FISA Section 702 directives, the NSA obtained email, chat, video, voice, photos, stored data, VoIP, file transfers, video conferencing, and login notifications from named U.S. service providers. The slides gave first-collection dates for Microsoft in 2007, Yahoo in 2008, Google and Facebook in 2009, YouTube in 2010, Skype and AOL in 2011, and Apple in 2012. The companies disputed the published implication of voluntary participation or direct NSA access to their servers; the government described a legally compelled provider process. That distinction changes the interface, not the custody: annual programmatic certifications replaced individualized target orders, and the largest American communications platforms became compelled collection points.

Boundless Informant, disclosed June 8, 2013, was the NSA’s metadata analytics dashboard — a tool that mapped and quantified the agency’s global surveillance reach by country in real time. March 2013 data showed the NSA had collected 97 billion pieces of intelligence globally in a thirty-day period, including three billion data elements from networks within the United States. This directly contradicted NSA Director James Clapper’s testimony to the Senate Intelligence Committee on March 12, 2013, that the NSA did not “wittingly” collect data on millions of Americans. The Boundless Informant data established that the collection was real, quantified, and ongoing — and that the agency’s director had offered false testimony to the Senate committee charged with overseeing it.

XKeyscore, disclosed July 31, 2013, was the NSA’s own self-described “widest reaching” collection system. Its training slides specified the data types captured: emails, chats, web-browsing traffic, pictures, documents, voice calls, webcam images, web searches, advertising analytics, social media activity, logged keystrokes, username and password pairs, file uploads, and Skype sessions. The system operated through more than 150 global sites with distributed servers processing collected data. An analyst could query the database in real time by name, email address, telephone number, or IP address with minimal bureaucratic authorization. This was surveillance capable of reconstructing an individual’s entire online behavioral history on analyst demand — a realized version of what LifeLog had described as an aspirational architecture.

MUSCULAR, disclosed October 30, 2013 by Barton Gellman and Ashkan Soltani at the Washington Post, reached a different level of the infrastructure entirely. Where PRISM operated through FISA-authorized directives to companies, MUSCULAR was a joint NSA-GCHQ program that tapped the private fiber optic cables connecting Google and Yahoo data centers to each other — the unencrypted internal communication links between company-owned servers traversing continents. The collection operated overseas, where U.S. legal constraints did not apply, and therefore required no court order. A top-secret NSA accounting dated January 9, 2013 showed field collectors had processed and transmitted back 181,280,466 new records in the preceding thirty days — audio, video, text, and metadata — at a volume that exceeded the agency’s analytical capacity. Google and Yahoo stated they were unaware of the program and encrypted their internal links after the disclosure. The encryption was a direct operational response to surveillance conducted without their knowledge against infrastructure they owned.

The upstream collection programs — BLARNEY, FAIRVIEW, STORMBREW, OAKSTAR — completed the picture by establishing that Section 702 authority extended to interception at the internet backbone itself, tapping the fiber optic cables and switching points at which internet traffic traverses U.S. territory, capturing data “about” targets rather than only “to” and “from” them, and thereby sweeping in vast quantities of communications unrelated to any surveillance target.

Five Eyes: Collection Without a National Border

The foundation of the intelligence-sharing arrangement now designated Five Eyes is a bilateral agreement between the United States and the United Kingdom dated March 5, 1946 — the British-U.S. Communication Intelligence Agreement, UKUSA — which formalized wartime SIGINT cooperation between the U.S. Army and Navy signals units and the British Government Code and Cypher School, GCHQ’s predecessor. The agreement was classified for sixty-four years and declassified by the NSA on June 24, 2010. Canada joined the arrangement in 1948, Australia and New Zealand in 1956. The designation “Five Eyes” (FVEY) emerged from internal intelligence community usage over subsequent decades.

Five Eyes is first a signals-intelligence collection and sharing alliance. Its shared infrastructure also creates routes around the practical and legal borders that would constrain a purely national system: one partner may collect traffic, host a sensor, or return intelligence that another could not acquire in the same manner at home. The Snowden record documents expansive sharing and partner access. The alliance’s architectural effect is clear: once collection, querying, and storage are distributed across five jurisdictions, the citizen’s legal boundary no longer matches the system’s technical boundary.

GCHQ’s TEMPORA program illustrated the mechanism at scale. The British program tapped fiber-optic cables landing in the United Kingdom and shared access with NSA; transatlantic infrastructure inevitably carried communications involving Americans as well as foreign targets. The disclosures establish collection and partner access at extraordinary volume. National statutes offer incomplete protection when the acquisition surface is global and an allied agency controls the cable landing point.

The arrangement with Israel’s Unit 8200 extended the sharing surface beyond Five Eyes. A memorandum published from the Snowden archive described transfer of raw NSA signals intelligence to Israel before NSA performed its own minimization, while requiring the Israeli side to handle, destroy, or protect U.S.-person communications under rules stated in the agreement. The document exposes permissive raw-data sharing; it does not support the simpler claim that the recipient obtained American communications free of any minimization duty. The risk lies in handing another service the first custody decision and relying on an agreement, rather than technical removal, to protect the people embedded in the stream.

Section 702 and the Expansion at RISAA

The legal architecture sustaining the domestic collection components has its roots in the Foreign Intelligence Surveillance Act of 1978, which established the Foreign Intelligence Surveillance Court as a secret judicial oversight mechanism for domestic intelligence surveillance and required individualized court orders targeting specific foreign agents. The architecture began eroding after September 2001. The President’s Surveillance Program, authorized by the Bush administration in the immediate aftermath and disclosed in the New York Times in December 2005, operated the NSA’s warrantless wiretapping entirely outside the FISA framework. The Protect America Act of August 2007 brought warrantless collection nominally under statutory authority by allowing surveillance of “persons reasonably believed to be outside the United States” without individualized court orders — the legal authorization under which PRISM began. The FISA Amendments Act of 2008 made the arrangement permanent and codified it as Section 702, while immunizing telecommunications companies for their cooperation in the preceding warrantless programs.

Section 702 as codified permits the Attorney General and Director of National Intelligence to jointly authorize targeting of non-U.S. persons reasonably believed to be located outside the United States, through annual programmatic certifications approved by the FISC rather than individualized orders. Companies receive legally binding directives compelling cooperation. The “incidental collection” of U.S. person communications, when those persons communicate with authorized foreign targets, is lawful under the statute and may be subsequently queried by FBI analysts without a warrant — the “backdoor search” mechanism that the Senate Intelligence Committee has documented as regularly exercised against American citizens.

The Reforming Intelligence and Securing America Act, signed into law as Pub. L. No. 118-49 on April 20, 2024, extended Section 702 authority through April 20, 2026 while simultaneously expanding it in ways the Electronic Frontier Foundation characterized as “the greatest expansion of surveillance authority in American history.” The expansion’s operative provision broadened the definition of “electronic communications service provider” beyond telecommunications companies and internet platforms to potentially encompass data centers, cloud service providers, equipment custodians, building managers, and any commercial entity that handles communications infrastructure in any form. Senator Ron Wyden described the provision as creating a mandatory assistance requirement for a category of businesses vast enough to encompass most of the commercial economy. The RISAA debate included debate over this provision that produced no clarifying limitation in the final text. The Congressional Research Service’s 2024 analysis (CRS Report R48592) identifies the expanded ECSP definition as the statute’s most legally consequential open question.

The reauthorization trajectory — 1978 FISA, the 2007 Protect America Act, 2008 Section 702, its 2012 and 2018 renewals, and the 2024 RISAA — shows a recurring conversion of scandal and reform pressure into renewed authority. Not every reform merely expanded surveillance: some imposed reporting, process, or program limits. The durable movement is still visible. In 2024, a warrant requirement for U.S.-person queries failed, the program survived, and the compelled-assistance universe widened through a disputed provider definition.

The Closed Loop and Its Implications for Targeted Populations

The architecture, taken as a whole, constitutes something the individual components conceal. In-Q-Tel seeds companies with classified procurement requirements embedded in commercial investment. DARPA funds foundational research whose civilian application creates dominant infrastructure platforms. FISA Section 702 compels those platforms to provide access to their data at programmatic scale. Five Eyes uses allied-nation collection to circumvent the domestic legal constraints that Section 702 still nominally requires. XKeyscore indexes the resulting data streams into a searchable behavioral database. Palantir integrates XKeyscore data into an analytical platform co-developed with intelligence analysts and now deployed to domestic law enforcement. The consumer internet is simultaneously an advertising and commerce platform and a surveillance infrastructure whose design reflects intelligence-community requirements embedded at the architectural level.

The implications for populations in active contact with this infrastructure — which is to say every person who uses a smartphone, a search engine, an email client, a financial platform, or a social media application — receive their most concrete treatment in the context of specific targeting programs. The Targeted Individuals and Neuroweapons literature documents a population whose reports of surveillance, stalking, and directed-energy effects are absorbed by the psychiatric system before the technical capacity question is examined. The architecture documented here establishes the surveillance substrate: a system that can, in principle, reconstruct the behavioral and associational history of any individual, in real time, from data already collected and retained — and can do so before any specific warrant or court order is sought, because the collection has already occurred under programmatic authority. The question the targeted individuals literature raises is not whether surveillance at this scale exists. The Snowden documents answer that question. The question is what proportion of the reports describe operation of a capability whose existence is now documented, rather than symptoms of a disorder whose diagnostic category was constructed without reference to the capability.

The Narrative Control architecture operates through the same infrastructure at the population level — not through individual targeting but through the aggregate behavioral data the system produces, which permits the kind of large-scale preference modeling and influence optimization that is documented in The Programmable Compliance Infrastructure. A platform that knows the precise behavioral signature of ideological drift, radicalization, or non-compliance in aggregate can surface content that stabilizes preferred distributions — not through censorship, which leaves a visible record, but through the quiet reweighting of information environments that leaves no discernible trace. The surveillance export economy through which these same capabilities are sold to authoritarian governments for deployment against their own populations is the global extension of the same integration: military-intelligence co-developed tools, entering commercial markets through exit events that preserve deniability, optimized for the monitoring and behavioral management of populations at scale.

Refusing the Apparatus

The architecture described here is not static. Its documented evolution — from FISA 1978 to Section 702 to RISAA 2024 — runs along a consistent vector: each successive legal iteration expands compelled assistance, narrows judicial oversight, and extends collection to new categories of infrastructure and provider. The civil liberties interventions that have succeeded — the Verizon order becoming a political catalyst, the PRISM disclosures forcing company encryption of internal links — have produced defensive adaptations that the subsequent legal cycle reabsorbs. The EFF’s characterization of RISAA as expansion rather than reform is empirically accurate when measured against the statute’s text.

The individual responses that remain available are, accordingly, partial and asymmetric. End-to-end encryption protects message content against MUSCULAR-type collection of unencrypted internal links; it does not protect against compelled PRISM-type access at the platform level, since platforms hold the decryption keys for stored data. Tor provides anonymization against IP-based identification; the NSA’s XKeyscore targets Tor users specifically, and correlation attacks against Tor traffic have been documented in academic literature as theoretically effective under sufficient adversarial vantage. The structural asymmetry is inherent to a system built with intelligence requirements embedded at the architectural level: defenses designed for the consumer threat model are insufficient against an adversary that helped design the infrastructure being defended against.

What the documentary record makes fully available — and what forensic engagement with that record provides — is an accurate map of the system’s actual architecture rather than its publicly maintained description. The publicly maintained description of the internet as a civilian infrastructure with limited and legally constrained government access is the description that the PRISM slides, the MUSCULAR accounting documents, the GCHQ TWO FACE wiki, and the In-Q-Tel Technology Access Agreements collectively disqualify. The accurate description is that the civilian internet and the surveillance infrastructure are the same infrastructure, built by partially overlapping institutions, governed by legal architecture that compels cooperation and rewards silence, and continuously extended at each legislative opportunity.

References

Barton Gellman and Laura Poitras. “U.S., British intelligence mining data from nine U.S. Internet companies in broad secret program.” The Washington Post, June 6, 2013. Primary report on the PRISM program, published simultaneously with The Guardian disclosure.

Barton Gellman and Ashkan Soltani. “NSA infiltrates links to Yahoo, Google data centers worldwide, Snowden documents say.” The Washington Post, October 30, 2013. Primary report on MUSCULAR; documents 181 million records collected in a single month. Available at washingtonpost.com.

Glenn Greenwald. “NSA collecting phone records of millions of Verizon customers daily.” The Guardian, June 5, 2013. Primary publication of FISA Court order Docket BR 13-80.

Glenn Greenwald and Ewen MacAskill. “NSA Prism program taps in to user data of Apple, Google and others.” The Guardian, June 6, 2013.

Glenn Greenwald. “XKeyscore: NSA tool collects ‘nearly everything a user does on the internet.’” The Guardian, July 31, 2013. Primary publication of XKeyscore training slides.

Sam Biddle. “Palantir Provides the Engine for Donald Trump’s Deportation Machine.” The Intercept, February 22, 2017. Includes publication of eight original Snowden documents on Palantir’s intelligence community integration, including the TWO FACE GCHQ documentation. Available at theintercept.com.

GCHQ VisWeek 2008 Conference Report. Published via The Intercept, February 22, 2017. Primary document characterizing Palantir as “sponsored by the CIA.”

NSA. “UKUSA Agreement Release, 1940–1956.” NSA official declassification, June 24, 2010. Includes the original March 5, 1946 British-U.S. Communication Intelligence Agreement. Available at nsa.gov/Helpful-Links/NSA-FOIA/Declassification-Transparency-Initiatives/Historical-Releases/UKUSA/.

Business Executives for National Security (BENS). Report of the Independent Panel on the CIA In-Q-Tel Venture. June 2001. The foundational review of the In-Q-Tel model, published two years after launch. Available at bens.org.

Defense Technical Information Center. ADA423535. “In-Q-Tel: CIA’s New Venture.” DoD paper on the CIA In-Q-Tel history and founding mechanism. Available at apps.dtic.mil.

In-Q-Tel. “In-Q-Tel Announces Strategic Investment in Keyhole.” Press release, June 25, 2003. Available at iqt.org.

Noah Shachtman. “Pentagon Kills LifeLog Project.” Wired, February 4, 2004. Contemporaneous primary report on LifeLog cancellation on the same date Facebook launched. Available at wired.com.

John M. Poindexter. Remarks at DARPATech 2002 Conference, Anaheim, CA, August 2, 2002. Full text archived by Federation of American Scientists at irp.fas.org/agency/dod/poindexter.html.

Michael G. Reed, Paul F. Syverson, David M. Goldschlag. “Anonymous Connections and Onion Routing.” 1997 IEEE Symposium on Security and Privacy. Naval Research Laboratory. DTIC accession ADA465335. Available at apps.dtic.mil.

Roger Dingledine, Nick Mathewson, Paul Syverson. “Tor: The Second-Generation Onion Router.” 2004. DTIC accession ADA465464. Available at apps.dtic.mil.

Electronic Frontier Foundation. “U.S. Senate and Biden Administration Shamefully Renew and Expand FISA Section 702.” April 20, 2024. Available at eff.org/deeplinks/2024/04/us-senate-and-biden-administration-shamefully-renew-and-expand-fisa-section-702-0.

Congressional Research Service. RISAA and Section 702: Legislative History and Key Provisions. CRS Report R48592, 2024. Available at congress.gov/crs-product/R48592.

FISA Court Opinion, September 17, 2024 (redacted). 2024 certifications under RISAA. Available at intelligence.gov.

Electronic Frontier Foundation. NSA Primary Sources Document Archive. Running compilation of primary Snowden documents with dates and source links. Available at eff.org/nsa-spying/nsadocs.