The rule no evidence can remove
A security control charges everyone, every day, for a loss that reaches a few. Nothing anyone can observe argues it out again, so the rules only accumulate.
· 5 min read
A security rule has a cost and a benefit, and they fall on different people at different rates. The cost lands on everyone the rule covers, every time they meet it. The benefit is at most the loss it prevents for the small share of them who would otherwise have been attacked. When that share is small, a rule can cost more than everything it could save while still stopping real attacks. It then stays, because its cost is never written down and its benefit can never be disproved. That is how stricter security becomes normal practice with a negative return, and it is an argument about where security effort goes, with no quarrel with security itself.
In 2009 Cormac Herley, of Microsoft Research, put numbers on the first half with a worked example: an exploit that hits 1% of users a year and costs each victim 10 hours to clean up.1 On his arithmetic, advice against it saves user time overall only if it costs each user less than 0.98 seconds a day.1 Valuing time at twice the US minimum wage across 180 million people online, he calculated that an hour a year of following some piece of advice would have to prevent $2.6 billion in direct losses to be worth it.1 He estimated that if users spent a minute a day reading web addresses to spot phishing sites, the time would cost about a hundred times more than all phishing losses.1
He also looked at who bore the losses. Banks at the time reimbursed most fraud, so a user who obeyed every rule saved almost nothing of their own.1 The usual view, he wrote, was that users are "hopelessly lazy"; his conclusion was that their rejection of security advice was "entirely rational", and that it was the security community that needed educating.1
The people paying for a rule are the first to see that it is not worth it, and their refusal to pay is read as the carelessness the rule exists to correct. Their dissent can then count as evidence for more rules.
Nothing comes off
Seven years later Herley explained why the list only grows.2 A system can be shown insecure by an attack that succeeds, but no observation can show it secure, so the claim that a measure is necessary can never be refuted.2 In his words, "the response to new information can only be to ratchet upward: newly observed or speculated attack capabilities can argue a countermeasure in, but no possible observation argues one out."2 Once a mistake is made, "we stay wrong and errors accumulate."2 He singled out measures with a recurring human cost, such as choosing and maintaining passwords, as the place where that waste is largest, and said he could find no organisation that backed its password guidance with evidence of better outcomes.2 He was careful about what that means: not that the measures have no value, only that nobody receives feedback on whether they work.2
So the cost is spread thin and unbooked, the benefit is a counterfactual, and the only events that ever move the rules are attacks, which can add a control and never subtract one. Each rule can be locally sensible while the pile has a negative return.
A rule that ate its own benefit
Forced password changes are a worse case, where the cost does not merely exceed the benefit but consumes it. Expiry is meant to lock out an attacker who already has your password.3 At the University of North Carolina every account's password had to change every three months, and in 2010 Yinqian Zhang, Fabian Monrose and Michael Reiter studied the password histories of more than 7,700 former accounts.3 Their method assumed that people make their next password by systematic edits to the current one, and it worked: given an old password, they could find the new one for 41% of accounts on average, in under three seconds each, and for 17% of accounts in fewer than five guesses at a login screen.3 They noted that, as far as they knew, nobody had measured whether expiry achieved its purpose before them, and concluded that their results call "into question the merit of continuing the practice."3
Every user paid in effort four times a year, and the effort went into a predictable edit that an attacker holding the old password could undo.
In 2017 the US National Institute of Standards and Technology wrote that verifiers "SHOULD NOT require memorized secrets to be changed arbitrarily (e.g., periodically)", while still requiring a change when there is evidence of compromise.4 It also advised against rules demanding a mix of character types, noting that their benefit was "not nearly as significant as initially thought" while the cost to usability was "severe", and that people meet such rules predictably, turning "password" into "Password1!".4
In April 2019 Microsoft published a draft of its recommended security settings for Windows that dropped expiry; the settings had asked for a change every 60 days, and once every 90.5 The post announcing it cited "recent scientific research" and called periodic expiry "an ancient and obsolete mitigation of very low value".5 Its reasoning was the arithmetic in miniature: expiry helps only if a password is stolen and then used within its lifetime, and anyone with evidence of a theft would act at once rather than wait.5 It pointed to banned-password lists and multi-factor authentication as better uses of the effort.5
What took the rule off, on this record, was measurement by people who could see both sides of the ledger, a university's own password histories and the bodies whose guidance others copy, rather than the complaints of the people paying. And the effort a negative-return rule consumes is effort that could have gone to a rule with a return, which is why the economics argues for security rather than against it.
Where it survives
The same Microsoft post explains why the rule outlives its evidence. Audits of these settings, it said, are usually scored by how many settings deviate from the recommended list, "How much red do we have on the chart?", and organisations "treat compliance numbers as more important than real-world security."5 A company with better defences that chose no expiry at all would be marked down anyway.5 One reader replied that compliance departments would keep forcing the same policies until the PCI and SOX rules, for payment cards and financial reporting, changed.5
An audit can see whether a rule is followed. It cannot see what the rule costs or whether it prevents anything, because neither is in the record. So the audit is where a measure goes on being necessary after the people who could measure it have said it is not.
Sources
- 1
- 2
- 3
- 4
- 5
About this piece
- Kind
- Argument. A reasoned case. The logic is the evidence; check it.
- Evidence
- Moderate
- Research
- Standard, 5 sources
- Made
- My idea. AI researched and wrote it, and an AI model checked each sourced claim against its source before I published it here.
- Updated
- Nov 21, 2026