Adobe, 2013: 153 million records and the world's worst crossword

Adobe's 2013 breach is not remembered for its size, although 153 million records was enormous for the time. It is remembered because the company had made a specific, fixable engineering decision about how to store passwords, and the dump turned that decision into a public demonstration of why it was wrong.

What happened

In late September 2013, Adobe found that intruders had been inside its systems. On 3 October it announced that customer information had been taken and that source code for several products had been copied, a combination that is unusual and, for a software vendor, worse than either on its own.

The numbers moved. Adobe first said 2.9 million customers, then raised it to 38 million active users. The file that circulated publicly held around 153 million records, including accounts that were long dormant or invalid. Both statements can be true at once: 38 million was Adobe's count of active accounts it considered affected; 153 million was what was actually in the database.

The password decision

A system that checks passwords does not need to be able to read them. That is the whole point of a password hash: the site stores a one-way transformation, and at login it repeats the transformation and compares. Done properly, a slow algorithm, a unique random salt per user, a stolen file yields almost nothing, and two users with the same password produce entirely different stored values.

Adobe did not hash. It encrypted, using Triple DES in ECB mode, with one key for everybody.

Two consequences follow, and they compound:

1. Encryption is reversible. A hash cannot be undone by anyone, including the company that made it. Encryption can, by anyone holding the key. The moment the key is compromised, every password in the file is readable.
2. ECB mode is deterministic. The same input produces the same output, every time, with no salt to break the pattern. So in the dump, every account that used 123456 shared an identical block of ciphertext. You could not read the password, but you could group together the hundreds of thousands of people who had chosen it.

The result was that researchers reconstructed the most common passwords in the file without ever attacking the encryption. The top entries were 123456, 123456789, password and adobe123: a ranking that has barely changed in the decade since.

The source code, which mattered more

Lost customer records damage the people in them. Lost source code damages everyone who runs the software. Adobe confirmed that code for ColdFusion and Acrobat, among others, was taken. ColdFusion in particular was widely deployed on internet-facing servers inside other organisations, and an attacker holding its source has a permanent head start in finding new vulnerabilities in it.

This is the part of the Adobe breach that is hardest to quantify and easiest to underrate. Customer data ages. Source code that ships into other people's infrastructure hands an attacker a research capability that keeps paying out for years.

The timeline

1. September 2013: Adobe detects the intrusion.
2. 3 October 2013: Public disclosure: 2.9 million customers, plus stolen source code.
3. 29 October 2013: The figure is revised to 38 million active users.
4. November 2013: A file of roughly 153 million records circulates; researchers begin reconstructing passwords from hints.
5. November 2016: Adobe settles with 15 US states and the District of Columbia for over $1 million, alongside an undisclosed consumer settlement.

What it changed

The Adobe file became a teaching artefact. Before it, "hash your passwords properly" was an argument security engineers had with product teams about theoretical risk. Afterwards there was a concrete, visual example of what the alternative looks like when the file gets out, and, crucially, one that a non-specialist could follow in ninety seconds.

It also killed the password hint as a feature. Hints existed to reduce support calls; they survive almost nowhere now, because Adobe demonstrated that a hint is a piece of plaintext that describes a secret, stored next to the secret, and available to anyone who takes the database.

The third legacy is more subtle. Adobe is one of the clearest cases of a breach figure that moved with the definition: 2.9 million, then 38 million, then 153 million, depending on whether you counted customers Adobe believed were affected, active accounts, or rows in the file. Almost every breach since has produced the same divergence, and reading a disclosure now means asking what exactly is being counted.

If you had an Adobe account in 2013