In October 2013, Adobe Systems disclosed a data breach that it initially characterized as affecting approximately 2.9 million customer records. Within weeks, the true scale became clear: the stolen database contained 153 million user records. It was, at the time, the largest breach of a single company ever disclosed. The breach contained three years of registration data for Adobe's software products - Creative Suite, Acrobat, ColdFusion, and the then-new Creative Cloud subscription service. What made it a landmark in security analysis was the password encryption scheme Adobe had used: a catastrophically weak approach that revealed itself through pattern analysis even before individual passwords were cracked.

The Adobe breach is not primarily remembered for its scale - subsequent breaches (Yahoo's 3 billion, LinkedIn's 117 million) eventually overshadowed it on that dimension. It is remembered because it became a teaching moment for the entire industry about the difference between password hashing and password encryption, and about what happens when password databases are stored in ways that preserve more structure than they should.

The Encryption Failure

Adobe did not hash its passwords - it encrypted them. This distinction matters enormously. Hashing is a one-way function: given a password hash, you cannot reverse it to recover the original password directly. The only attack is to guess passwords and compare hashes. Encryption is a two-way function: given the encrypted password and the encryption key, you can decrypt directly to the original. Adobe stored its passwords encrypted with 3DES (Triple DES) in ECB mode.

ECB (Electronic Codebook) mode is the simplest and most dangerous encryption mode for this use case. In ECB mode, identical plaintext blocks produce identical ciphertext blocks. This means that identical passwords produce identical encrypted values. The same problem that affects unsalted SHA-1 (as in LinkedIn) applies to 3DES-ECB: if two users have the same password, their encrypted passwords in the database are identical.

But Adobe's failure went further than ECB mode. The structure revealed by the encrypted values allowed sophisticated analysis before any password was even cracked. Security researchers Jeremi Gosney and Troy Hunt performed a famous analysis using the password hints stored in the database (Adobe stored unencrypted password hints alongside the encrypted passwords). By grouping encrypted values and examining the associated hints, patterns emerged immediately. All users who had the identical encrypted value and hints like "the usual" or "adobe" or "same as always" revealed that common encrypted values corresponded to common passwords. Cross-referencing encrypted values across users allowed construction of frequency tables that mapped directly to password popularity.

[TECHNICAL NOTE]
The Adobe ECB mode failure illustrated through cryptographic analysis: in ECB mode, a 64-bit plaintext block always maps to the same 64-bit ciphertext block under the same key. For 3DES with a plaintext password shorter than 8 characters (padded to a block boundary), the entire password fits in one block - meaning the entire encrypted value is deterministic. For passwords longer than 8 characters, each 8-byte block encrypts independently. This means two passwords that share the same first 8 characters have the same first encrypted block, even if they differ afterward. The correct approach was to use a hashing algorithm designed for password storage (bcrypt, scrypt, Argon2) with per-user random salts. The 3DES-ECB approach was not just weak - it was strictly worse than plaintext storage in some ways, because it gave users false confidence while providing nearly no additional protection once the key was obtained. The encryption key itself was stored on Adobe servers - obtaining the database almost certainly meant obtaining the key too, since they resided on the same infrastructure.

Crossword-Puzzle Cracking

The combination of encrypted passwords, password hints, and the frequency-grouping afforded by ECB mode enabled a novel attack methodology that journalists described as "crossword-puzzle cracking." Because identical passwords produced identical encrypted values, a researcher who cracked one instance of a given encrypted value had cracked all instances.

Troy Hunt's analysis of a 3.8 million record subset (the portion that had leaked publicly) found that the most common encrypted value appeared 1,911,938 times. If that value was cracked (or guessed from hints), nearly 2 million passwords were recovered simultaneously. The hint analysis revealed that value corresponded to "123456". The second most common value appeared 446,162 times. And so on. The structure of the database essentially made it a frequency analysis problem rather than a per-password cracking problem.

Researchers constructed a "cheat sheet" that allowed anyone with access to the database to look up many passwords directly by matching the encrypted value to a known plaintext - without any cryptographic attack at all, just frequency analysis and hint matching.

[WARNING]
The Adobe breach's other content amplified its harm beyond the password exposure. The database also contained source code for several Adobe products, including the backend of Adobe ColdFusion and portions of Adobe Acrobat. Source code theft is serious because it enables an attacker to analyze the code for vulnerabilities that would be much harder to find through external testing. ColdFusion was a particularly concerning loss: it is widely used in web application development and is often deployed in enterprise environments. Source code analysis would enable zero-day vulnerability discovery with much greater efficiency than external fuzzing or testing. Adobe and the security community expected (and several were found) vulnerabilities to emerge in Adobe products following the source code disclosure. The source code theft was in some ways more strategically significant than the credential database - credentials are perishable (changed over time), but architectural knowledge from source code provides persistent intelligence for finding new vulnerabilities.

Breach Vector and Attribution

The breach was attributed to the same attacker (or group) who had breached several other companies around the same time - the actor was posting stolen databases on a hosting service called "Anon" and appeared to be collecting large credential databases rather than immediately monetizing them. The Adobe breach was discovered by security journalist Brian Krebs and researcher Alex Holden, who found the stolen data on a server used by a group that had also breached multiple other organizations.

The initial access vector into Adobe was not definitively established in public disclosures. Adobe's systems ran ColdFusion (its own product) for web applications, and ColdFusion had a history of critical vulnerabilities. At the time of the breach, active ColdFusion zero-days were circulating. The FBI investigated and Adobe cooperated, but no arrests resulted from the investigation.

Adobe notified affected customers by email in October 2013 and forced password resets. The company settled a class action lawsuit for an undisclosed amount in 2014. The FTC did not pursue enforcement action. The California Attorney General's office opened an investigation but did not publicly announce enforcement results. By the standards of 2013, the breach was enormous but regulatory and financial consequences were modest.

[IOC]
Adobe 2013 breach summary: disclosed October 3, 2013. Initial disclosure: 2.9 million records. Actual scope: approximately 153 million user records. Data included: email addresses, encrypted passwords (3DES-ECB), password hints (unencrypted), internal employee IDs, customer names. Source code also stolen: Adobe ColdFusion, Adobe Acrobat, portions of Adobe Reader. Password storage: 3DES (Triple DES) in ECB mode - neither a hash nor a properly implemented encryption scheme for password storage. The encryption key was likely stored on the same infrastructure as the database, making encryption security illusory once the database was obtained. Most common cracked password: "123456" (nearly 2 million instances). Brian Krebs and Alex Holden discovered the stolen data on an open directory on a server hosting data from multiple breaches. The server also contained data from other major breaches including PR Newswire, several banks, and government contractors. No public attribution or arrests. Adobe paid to settle class action suits in 2014 for undisclosed amount. Password hint analysis combined with ECB-mode frequency analysis enabled credential recovery without per-password cryptographic attack - a novel breach analysis technique that emerged from this incident. Notable downstream effect: Adobe breach credentials became a common component of credential stuffing lists used against other services for years afterward.