On March 7, 2017, WikiLeaks began publishing what it called "Vault 7" - a collection of CIA hacking tools, techniques, and documentation that represented the largest unauthorized disclosure of classified intelligence material in the CIA's history. The first release alone contained 8,761 documents and files. Subsequent releases continued for months. The disclosed material included working exploit code, detailed technical documentation of attacks against iOS, Android, Windows, Samsung smart TVs, and network infrastructure, and internal CIA operational security guidelines that revealed how the agency tried to avoid attribution.
The source was eventually identified as Joshua Adam Schulte, a CIA software engineer who had worked in the Engineering Development Group - the unit responsible for developing the tools. Schulte was convicted in 2022 on all nine espionage counts after a first trial ended in a hung jury. The Vault 7 leak was operationally damaging to the CIA in ways that may not be fully known publicly, as tools that take years to develop were rendered useless once their signatures appeared in the public domain.
What Was Disclosed
The Vault 7 Year Zero release documented the CIA's use of zero-day vulnerabilities and exploits across a wide range of platforms. iOS exploits allowed persistent access to iPhones. Android exploits bypassed encryption. Windows exploits included tools that could be stored on removable drives and execute automatically. "Weeping Angel" was a tool for Samsung smart televisions that could place the TV in a "fake off" mode - appearing to be powered down while actually recording audio from the room and transmitting it over the internet.
"Marble Framework" was a tool for obfuscating CIA malware's source code to defeat forensic analysis and misdirect attribution - strings could be inserted in foreign languages (Russian, Chinese, Arabic, Korean, Persian) to implicate other nation-states in CIA operations. The disclosure of Marble was directly relevant to attribution debates about previous attacks: if the CIA was using Marble to insert foreign-language strings in its malware, then foreign-language strings in malware were less reliable as attribution indicators than the security community had assumed.
Joshua Schulte and the Investigation
FBI investigation identified Joshua Schulte through digital forensics of the CIA's internal development systems. Schulte had administrative access to the Confluence wiki and file server from which the Vault 7 material was taken. The investigation found that he had exploited a CIA development server to create a snapshot of the relevant data, reverse-revoked his own access to cover his tracks, and transmitted the data to WikiLeaks.
Schulte had left the CIA in 2016 following a workplace conflict with a colleague that had escalated into formal complaints and investigations. Prosecutors argued the leak was motivated by grievance. At his 2022 trial, Schulte represented himself in part, cross-examining CIA and FBI witnesses at length. He was convicted on all nine espionage counts, one count of making false statements to the FBI, and one count of obstruction of justice. He was sentenced to 40 years in federal prison in February 2024.
The Attribution Implications
Beyond the operational exposure, Vault 7's disclosure of the Marble Framework had lasting implications for how the security community approaches malware attribution. Before Vault 7, the presence of specific language strings, coding patterns, and operational security choices in malware code was used as attribution evidence - Chinese APT groups had consistent code patterns, Russian groups had Russian-language strings in error messages, and so forth.
Marble demonstrated that at least one major intelligence agency was deliberately inserting fake attribution markers into its tools. If the CIA did this, the reasonable inference is that other sophisticated intelligence agencies did too. This doesn't invalidate technical attribution entirely - technical indicators are still meaningful when corroborated by infrastructure analysis, targeting patterns, and timing - but it added a layer of uncertainty to single-indicator attribution claims and reinforced the principle that confident attribution requires convergent evidence from multiple independent sources.