Most of the security advice circulating in offices is twenty years old and has been explicitly revised by the bodies that issued it. We keep changing passwords every ninety days and adding an exclamation mark at the end, while the guidelines we’re supposed to follow say to stop doing that.
This guide starts there: first what changed, then the few measures that actually move the risk, in order of effectiveness.
The password rules changed (and almost nobody knows)
The reference document is NIST SP 800-63B, the Digital Identity Guidelines, which in its recent revisions recommends that service providers:
- not impose periodic password expiry absent evidence of compromise;
- not impose composition rules of the “at least one capital, one number and one symbol” kind;
- check passwords against lists of already-compromised credentials;
- favour length, allowing long passphrases.
The reason is behavioural and documented: forcing frequent changes produces
predictable variations of the same password (Summer2025! → Summer2026!),
which reduces security rather than increasing it. Composition rules have the
same effect: they push people towards patterns that attack tools know
extremely well.
You’ll find the same position, in different words, in the guidance from the UK’s NCSC, which recommends passphrases made of random words.
What follows in practice:
| Old rule | What the guidelines say now |
|---|---|
| Change it every 90 days | Change it when there’s a reason: suspected compromise, breach notification |
| At least one symbol and one number | Length is what counts; four random words beat P@ssw0rd! |
| Never write it down | A password manager beats memory, because it makes a different credential per service possible |
| One strong password is enough | It isn’t: it has to be different on every site |
The last point is the important one. Reuse is the mechanism that makes breaches profitable: credentials stolen from one site are automatically tried against dozens of others. One very strong reused password is a problem; twenty mediocre but all-different passwords held in a manager is a much better position.
Not all second factors are equivalent
This is the second thing generic articles get wrong: they present “turn on two-factor authentication” as a single checkbox. It isn’t.
| Second factor | Phishing resistance | Notes |
|---|---|---|
| SMS code | Low | Interceptable and transferable; discouraged as the primary method |
| App code (TOTP) | Medium | Much better than SMS, but a fake site can still ask you for the code |
| Push notification with approval | Medium | Vulnerable to notification fatigue: approved by reflex |
| FIDO2 key / passkey | High | The credential is bound to the domain: on a fake site it simply doesn’t work |
The technical reason the last row differs: with FIDO2 the device verifies which site it’s responding to. If the domain doesn’t match, it produces nothing worth stealing. With a six-digit code, an attacker only has to ask for it within the validity window and relay it to the real site. That’s how “adversary in the middle” phishing kits work, as described in CISA material on phishing-resistant MFA.
The non-ideological conclusion: any second factor beats none, but on the accounts that really matter (primary email, banking, corporate access) it’s worth moving to passkeys or a hardware key.
Phishing, recognised by structure rather than by typos
The classic advice, such as “look for spelling mistakes” or “beware of attachments”, has aged badly: credible messages are now the norm. The more robust criterion isn’t linguistic but structural.
Three questions, in order:
- Does the message create urgency or secrecy? “By today”, “don’t mention this to anyone”, “the payment will fail”. Haste exists to prevent verification: it’s the most consistent signature.
- Is it asking me to authenticate starting from a link? That’s where credentials get lost. The rule is simple: you reach a login through your own bookmarks, never through the message.
- Is it asking for a change to money or data? New bank details for a supplier, an out-of-process transfer, sending a list of employees. Here verification goes through a different channel, using a number you already had, not the one written in the email.
The third case has a name, business email compromise, and requires no malware at all: just a plausible request made to the right person. It’s why the annual ENISA threat reports consistently place social engineering and credential theft among the leading threats, alongside ransomware.
The few measures that actually move the risk
In order of effect-to-effort.
- Update your software. Tedious and decisive: a very large share of successful attacks exploits vulnerabilities for which a fix already existed.
- Use a password manager. It solves reuse, which is the real problem.
- Turn on the best available second factor on critical accounts.
- Enable disk encryption on laptop and phone. It turns a hardware theft into an inconvenience rather than a data breach.
- Check your addresses against known breaches, for example with Have I Been Pwned, and change the passwords for the services involved.
- Separate your uses. The personal account doesn’t touch company systems and vice versa. That applies to AI tools too: what can and can’t be pasted in is covered in how to use ChatGPT at work.
- On the move, use your phone’s connection rather than open Wi‑Fi. It’s the simplest measure for people working off-site, along with the others in remote work: what actually works.
What to do in the first ten minutes, if it happened
The moment when most damage is done is the one right after, out of embarrassment.
- Delete nothing. The message, the log, the email: they’re needed to work out what happened.
- Report immediately, even if you think it was your fault. A false alarm costs nothing; an hour of silence can cost a great deal.
- Change the password for the service involved and invalidate active sessions where the service allows it.
- Disconnect the machine from the network if you suspect code execution, but don’t power it off unless asked: shutting down can destroy useful evidence.
There’s also a regulatory angle that concerns the company rather than you personally, but it’s useful to know: in the case of a personal data breach the GDPR requires notification to the supervisory authority without undue delay and, where feasible, within 72 hours. That’s why reporting immediately isn’t zeal: it’s what allows the organisation to meet a legal deadline.
For companies above certain thresholds and in specific sectors, the NIS2 directive (EU 2022/2555) adds risk-management and reporting obligations: if you work in energy, health, transport, water or digital services, it likely applies to your employer.
My take: the weak link is usually the process
I find “the user is the weakest link” a convenient and largely false statement. If an organisation has designed a process where one person, with one email, can authorise a payment without independent verification, the problem isn’t the person: it’s the process. Security that works is security that assumes human error and places a control afterwards, not security that asks people never to make mistakes.
The less diplomatic part: mandatory annual training with a final quiz mostly produces a certificate. What changes behaviour is a list of two or three absolute rules, repeated until they’re automatic: “you reach a login through your bookmarks”, “bank detail changes get verified by phone on an old number”.
Frequently asked questions
Do I really have to change my password every three months? Current NIST guidance advises against periodic expiry absent evidence of compromise, because it induces predictable variations. If your employer imposes it, that’s an internal policy: you can comply using a manager that generates a new, unrelated password each time.
Is a password manager safe? It’s a single point of failure. It’s a concentrated risk, but the comparison has to be with the realistic alternative, which for most people is reusing the same password everywhere. With a strong passphrase and a second factor on the manager itself, the balance is clearly favourable.
Do passkeys replace passwords? On services that support them, yes: they’re bound to the domain and therefore can’t be replayed on a fake site. The transition is gradual, so they’ll coexist with passwords for a while.
My computer is company-issued: isn’t security IT’s problem? The infrastructure is; the daily decisions aren’t. No technical control stops an authorised person sending a file to the wrong recipient or approving a payment.
Is it worth specialising in cybersecurity? It’s among the areas with steadily growing demand and a declared shortage of staff in European surveys. The demand picture is in which jobs are most in demand and what skills employers ask for.
Sources
- NIST, SP 800-63B, Digital Identity Guidelines: recommendations on password length, dropping periodic expiry and composition rules, and checking against compromised-credential lists.
- CISA. Guidance on phishing-resistant multi-factor authentication and the distinction between phishable and non-phishable factors.
- ENISA, Threat Landscape: annual report on leading threats, including ransomware, social engineering and threats against credentials.
- NCSC. Password and passphrase guidance for organisations and individuals.
- EU Regulation 2016/679 (GDPR). Notification of personal data breaches to the supervisory authority within 72 hours where feasible.
- Directive (EU) 2022/2555 (NIS2). Risk-management and incident-reporting obligations for the listed sectors.
- Have I Been Pwned. Public service for checking whether an email address appears in known breaches.



