Your Passwords Should Be Shorter
Where
PasswordsCon
BSides Las Vegas
Duration
31 min
Works through the math of password length and hash cracking to argue that 15 random characters can be enough.
Summary
The talk opens with a joke about hacker-conference one-upmanship over password length, then makes a technical point. The bcrypt hashing function only uses the first 72 bytes of input, and multi-byte UTF-8 characters such as emoji use up that budget faster, so very long passwords can be silently cut short. Past vendor storage failures are cited as a reason not to assume large companies get this right.
To judge how long a password really needs to be, the speaker builds a threat model. The worst case is a fast MD5 hash, attacked by a cloud GPU cluster costing roughly a billion dollars a year, with a fudge factor added on top. Over an eight-year lifetime, that works out to about 85 bits of entropy, which 15 random letters and digits supply. Slower hashes such as bcrypt, scrypt, PBKDF2 and Argon2 widen the margin considerably, and a hardware security key is presented as a stronger defense against phishing and account takeover.
In the question period, the speaker concedes that diceware-style passphrases are a different case because people can memorize them, and notes that a password manager vendor chose a similar default length for usability reasons. Audience members warn that advice like this gets pushed to extremes, and suggest standards bodies as a channel. The takeaway: generate 15-character random passwords, and add a special character only where a system insists on one.
Related talks
PasswordsCon·21 min
Travis Moore
Explains how botnets generate high-entropy synthetic passwords to flood logins and evade breach-based detection.
PasswordsCon·55 min
Filipi Pires
Shows how one overly broad cloud permission can let an attacker escalate privileges across an entire organization.
PasswordsCon·39 min
Filipi Pires
Walks through cloud identity misconfigurations and shows how graph-mode tooling maps privilege escalation paths.


