Vibe-Coding Your HR Software: 9 Concrete Risks You Should Know
Article summary
- AI-generated code has been shown to reproduce known vulnerability patterns (hardcoded credentials, broken authentication, unsanitized inputs) at a higher rate than code built with a security review process.
- A vibe-coded HRIS typically ships with no documented data residency, no role-based access control, and no audit trail of who accessed what.
- Candidate data (resumes, background checks, voluntarily disclosed health or immigration details) is often more exposed than employee data in DIY-built systems, since applicants have no ongoing relationship with the company to fall back on.
- Compliance obligations like PIPEDA and provincial privacy law aren’t retrofitted easily onto software that wasn’t built with them in mind from day one.
- When something breaks, there’s no vendor SLA, no support line, and often no one left on the team who fully understands the code.
AI is transforming the world of work as we know it, so much so that we now hear companies saying: “Honestly, we were thinking of just vibe-coding something ourselves.”
And honestly, we get the appeal. You open an AI coding tool, describe what you want, and twenty minutes later you’ve got something that looks like an HR system. No vendor contract, no onboarding calls, no monthly invoice. For a founder or an office manager juggling fifteen things before lunch, that’s definitely a seductive shortcut.
Here’s the problem: an HRIS is the system holding your employees’ SIN numbers, salaries, medical accommodation requests, and your candidates’ resumes and background checks. When that system fails, it fails in the form of a data breach, a compliance violation, or a very uncomfortable conversation with someone whose personal information just leaked.
So let’s walk through, honestly, everything that can go wrong when you vibe code your HRIS.
What “vibe-coding your HRIS” actually means
Vibe-coding is building software by describing what you want to an AI tool and accepting what it generates, with little to no manual review of the underlying code. It’s great for prototypes, internal scripts, or a generic app nobody depends on. It’s different when the output is handling personal data for every employee and candidate in your company.
The Concrete Risks of Vibe-Coding Your HR Software
1. The security wasn’t designed in
Real HR software goes through security architecture decisions before a single feature gets built: how data is encrypted at rest and in transit, how sessions are managed, how permissions are scoped. An AI tool generating code on the fly isn’t making those decisions deliberately. It’s producing plausible-looking code, and “plausible-looking” and “secure” are not the same thing.
Security researchers have repeatedly found that AI-generated code reproduces well-known vulnerability patterns (hardcoded credentials, broken authentication, unsanitized inputs) at a meaningfully higher rate than code written by developers following a security review process.
2. You don’t actually know where the data lives
Ask yourself: if you vibe-coded your HR database tomorrow, could you say with certainty which cloud provider is hosting it, which country the servers are in, and who else has infrastructure-level access? Most people building this way can’t answer that — because the AI tool chose the defaults, and the defaults are rarely built with compliance in mind. Data sovereignty isn’t a nice-to-have for employee records, it’s a question you need a real answer to.
3. No one’s checking who has access
In a proper HRIS, access is role-based: a manager sees their team’s data, payroll sees compensation, and nobody sees medical notes unless they absolutely need to. Building this correctly is tedious, unglamorous work: exactly the kind of thing that gets quietly skipped when you’re optimizing for “does it technically run.” The result, more often than not, is a system where access control is an afterthought, or doesn’t exist at all.
4. Compliance isn’t a feature you bolt on later
Whether it’s PIPEDA at the federal level, provincial privacy laws, or sector-specific retention requirements for payroll and tax records, HR data comes with legal obligations attached. A vendor that’s done this seriously can show you a SOC 2 Type II certification and explain exactly how they meet these obligations. A system someone vibe-coded over a weekend has none of that, and retrofitting compliance onto software that wasn’t built for it is far harder than building it in from day one.
5. Candidate data might be the riskiest part of all
It’s tempting to think about this purely in terms of current employees, but your recruitment pipeline is arguably more exposed. Resumes often contain more personal detail than an employee file ever will (addresses, phone numbers, sometimes immigration status or health information disclosed voluntarily). That data touches your system before you’ve even decided to hire someone, which means a breach there affects people who never agreed to be your employees and have no ongoing relationship with your company to fall back on.
6. No audit trail means no answers when something goes wrong
If an employee’s salary gets viewed by someone who shouldn’t have seen it, can you trace exactly who accessed what, and when? In a real HRIS, that’s a standard logging feature. In a vibe-coded one, it’s probably not there — which means that if something does go wrong, you won’t just have a breach, you’ll have a breach you can’t even fully investigate.
7. “AI wrote it” is not the same as “someone tested it”
There’s an assumption floating around that because an AI generated the code, it’s inherently more reliable than something a person could have built. It’s the opposite problem: nobody wrote it with intent, and — critically — nobody is accountable for having reviewed it.
A human developer who ships a vulnerability can be asked what they were thinking. An AI tool can’t be held to that same standard, and often nobody on your team has the security background to catch what it missed.
8. There’s no one to call when it breaks
When something goes wrong with vendor software, you open a support ticket. When something goes wrong with code an AI generated for you last spring, you’re on your own — or worse, the one person who understood how it was duct-taped together has since left the company.
This is the technical debt problem hiding behind every “we’ll just build it ourselves”: the system works fine right up until it doesn’t, and then there’s no SLA, no support line, and no one with context to fix it quickly.
9. The real cost shows up later, not now
Vibe-coding looks free in month one. The real cost — a breach, a compliance fine, weeks of engineering time to patch a system nobody fully understands, or a rebuild from scratch once you realize the risk shows up much later.
What to do instead
None of this means you need the most expensive enterprise platform on the market. It means the bar for “software touching personal employee and candidate data” should be higher than “software for tracking inventory.” At minimum, look for a vendor that can show you independently verified security certifications, a clear answer on data residency, and role-based access control out of the box.
Frequently Asked Questions
Is it ever okay to vibe-code internal HR tools?
For something low-stakes and data-free — like a shift swap request board with no personal information attached — the risk is low. The moment personal data (SIN numbers, salaries, health information, background checks) enters the picture, the risk profile changes entirely.
What's the single biggest risk of a vibe-coded HRIS?
Lack of accountability. There’s no certification, no audit trail, and usually no one on the team with the security expertise to catch what’s missing, which means problems tend to surface only after a breach has already happened.
How is this different from using no-code tools like Airtable or Notion for HR?
It’s a matter of degree, not kind — generic no-code tools weren’t built for HR-specific compliance needs either. But AI-generated custom code adds an extra layer of risk: nobody, including the person who “built” it, fully understands or has reviewed what’s actually running under the hood.
Curious to learn more?