In a business of ten people, almost every incident involving staff is someone trying to do their job quickly. Controls built for malice will not stop it.
Short answer: Mostly, yes, at least at small business scale. "Insider threat" is a real category and deliberate insiders exist, but in a firm of three to twenty five people nearly every incident involving a staff member is someone taking a shortcut to finish work on time. Treating that as a threat leads to controls that punish competence and miss the actual risk.
Key takeaways
In security literature, insider threat covers anyone with legitimate access who causes harm. It is usually split three ways: the malicious insider who acts deliberately, the negligent insider who ignores policy, and the compromised insider whose account is being used by somebody else.
Each is real. But bundling them produces a category so broad it stops guiding decisions, because the three demand opposite responses. The malicious insider calls for detection and evidence. The compromised insider calls for authentication and anomaly detection. The person taking a shortcut calls for something else entirely, and that is the one you will actually meet.
The Australian regulator's breach statistics have shown human error as a material share of notifications in every reporting period since the scheme began, alongside malicious and criminal attacks. The specific errors are mundane and familiar: information sent to the wrong recipient, an attachment containing more than intended, a document shared with a link set too broadly.
Add the newer version, which does not yet appear cleanly in any published category: a bookkeeper pasting a client's bank statement into a chat tool to summarise it, an engineer uploading a tender document to get help with the methodology, a developer dropping a connection string into a chat window while debugging.
Look at who does these things. They are your most productive people, working on your most sensitive matters, under time pressure. Bookkeeping, legal drafting, tender writing and engineering are exactly the work AI helps with most, and exactly the work whose documents contain client identifiers, pricing and credentials. The exposure concentrates where the competence is.
Calling that a threat gets the situation backwards. Nobody in that story is working against the business. They are working for it, slightly too fast.
You buy surveillance. Insider threat tooling is built to gather evidence: keystrokes, screenshots, content logging, behavioural scoring. In a ten person office that is both disproportionate and corrosive. Everyone knows within a week.
You lose your reporting channel. This is the expensive one. If staff believe the tool exists to catch them, nobody tells you when they paste something by accident. You find out months later, or never. In jurisdictions with a breach notification duty, the clock starts when you become aware, and a workforce that has stopped telling you means you become aware last.
You solve for the rare case. Controls designed to catch a determined insider are poor at preventing an ordinary mistake, because the mistake happens in a second and involves no concealment. You end up with heavy machinery pointed at the wrong event.
You get compliance theatre. People find the path of least resistance. Block the tool on work machines and it moves to phones, where you have no relationship with the device at all.
Assume good faith and design for speed. Three principles.
Show, do not block. The person holds the context you do not. They know whether that account number belongs to a client or is a made up example. A control that informs them and lets them decide preserves the judgement; a control that blocks trains them to route around it.
Aggregate, do not read. An owner needs to know that financial details were nearly sent to a chat tool three times last month, and by which team. They do not need to read anyone's prompts, and being unable to is a feature, not a limitation.
Make reporting safe and say so. State plainly that reporting an accidental paste is expected and will not be punished, and then honour it the first time it is tested. That first response sets whether you hear about the second one.
This is how Prinn is built, deliberately. It runs on each person's computer and checks a prompt before it is sent. If it finds something sensitive it shows what it found and offers three choices: redact it, stop, or send it as is. The person decides, and it never blocks anyone. Prompt content never leaves the machine, including to us, so there is no prompt log for anyone to read. An owner sees the category, the tool, the date, the device and the choice made.
It is reliable on credentials, financial details and identity numbers. Names, home addresses and dates of birth written into ordinary prose are harder, and it is still improving at those.
Do not overcorrect. There are situations where deliberate insider risk is the real concern: a contested departure to a competitor, a dispute over client ownership, a role with unsupervised access to client funds. In those cases access reviews, segregation of duties and proper offboarding are the right instruments, and this page is not an argument against them.
The point is narrower. Those are the exceptions, and building your everyday posture around them costs you the trust that makes the everyday case manageable.
Ask one question about any control you are considering: does this make it easier or harder for a member of staff to tell me they made a mistake?
Anything that makes it harder is buying you less than it costs. In a business without an IT function, the staff are not the perimeter to defend against. They are the only sensors you have.