When AI Lies: Why Polished Applications Need Stronger Verification

TL;DR

• The first AI risk most leaders see is internal: bad summaries, confident errors, and weak human review.
• The next recruitment risk is external: AI can help applicants smooth gaps, rewrite experience, and present a story that feels more credible than the facts support.
• The bigger problem is fake coherence: a version of someone that appears consistent across CV, interview answers, references, and supporting material.
• Traditional checks remain necessary, but they do not always test a polished digital story.
• In high-trust roles, the answer is stronger verification and proportionate, human-led digital due diligence.

What people mean when they say AI lies

One of the recurring points during the session was that AI does not “lie” in the human sense. It predicts. It fills gaps. It gives you the answer that seems most likely to fit the pattern.

Exactly why it creates so much false confidence.

The danger is not that AI sounds obviously wrong. The danger is that it sounds professionally right.

A weak answer written badly is easy to challenge. A weak answer written fluently is much harder. Once the wording sounds clean, structured, and authoritative, people start treating it as judgment rather than output.

That is risky enough when the AI is inside your organisation - drafting communications, summarising meetings, or supporting decisions.

It gets more interesting when the person using AI is outside your organisation and trying to get in.

The recruitment problem is different

The recruitment risk goes beyond fake documents or obvious fabrication.

The practical change is that AI helps people remove the rough edges that used to trigger further scrutiny.

It can help rewrite employment history into cleaner language. It can help turn vague experience into confident-sounding capability. It can help explain away awkward gaps, standardise tone across documents, improve personal statements, and prepare calm answers to likely interview questions.

The result is that AI can help a weak or misleading application feel more complete than it really is.

That creates pressure because recruitment already depends heavily on trust. Employers are making decisions based on what has been declared, what has been evidenced, and what appears credible in the round.

If AI makes that round picture look smoother, the burden on the employer changes.

During that session, we ran two LLM prompts live. They were close enough to feel like the same exercise, but they produced different outputs, including the model confidently getting Frank’s name wrong.

That is the mechanism to watch in recruitment. If a tool can produce a polished account that sounds authoritative while getting basic facts wrong, it can also help someone create a smoother version of employment history, references, or capability than the evidence supports.

The problem is fake coherence

This is the part I think more leaders need to focus on.

Historically, deception often carried friction. Timelines clashed. Language changed between documents. A covering note felt different from the CV. A reference sounded oddly thin. An answer in the interview did not quite match the written version.

AI removes some of those tells. It does not just generate content. It generates coherence.

It helps create consistency across the full candidate story. The CV sounds polished. The personal statement sounds measured. The interview answers are rehearsed. The explanation for a gap feels controlled. The whole thing hangs together.

That does not prove dishonesty. Plenty of candidates will use AI in entirely ordinary ways.

The governance problem is different: coherence is no longer especially reassuring on its own.

When the story looks perfect, the useful question becomes whether it has been verified properly.

Why this matters more in high-trust roles

This issue is more acute in roles involving children, vulnerable adults, access to sensitive systems, financial authority, reputational exposure, or positions where professional judgment carries serious consequence.

In those settings, a bad hire rarely stays confined to HR.

It becomes a safeguarding issue. A governance issue. A security issue. Sometimes a regulatory issue. Very quickly, it becomes a leadership issue.

That is why the standard cannot stop at “the application looked good” or “nothing obvious came up”.

High-trust recruitment needs a process that can stand up after the fact – especially when somebody asks what was checked, what was relied on, and why the decision felt reasonable at the time.

Why traditional checks will not always solve it

DBS checks, right-to-work checks, qualification checks, and references remain essential.

They were never designed to do every part of this job.

A DBS check can confirm recorded information. A reference can confirm what is disclosed. A qualification check can confirm whether a credential exists.

None of that automatically tells you whether the broader story around a person has been professionally polished in a way that conceals risk, inconsistency, or relevant context.

This is where many organisations drift into manual online searching and call it due diligence.

That usually creates a second problem.

Ad hoc searching is inconsistent. It is hard to audit. It is easy to overreach. Two managers can look at the same role, search in different ways, and come away with different conclusions.

That leaves too much to chance.

What a stronger response looks like

A stronger response needs structure, proportionality, and clear ownership.

It starts with a few practical decisions:

• decide which roles justify deeper due diligence
• define who is allowed to carry it out
• tighten verification around employment history, qualifications, references, and unexplained inconsistencies
• treat polished application material as a claim to be tested, not evidence on its own
• make sure escalation, documentation, and final judgment stay with trained humans

Where the role justifies it, there is a case for an additional layer of digital due diligence.

In some high-trust environments, organisations are starting to add structured digital due diligence alongside traditional screening, not to replace human judgment, but to strengthen visibility where conventional checks were never designed to operate.

At Safehire, that is the thinking behind our Digital Risk Screening.  It sits alongside traditional checks, adding a lawful, proportionate, and governed layer that helps organisations surface relevant digital signals which conventional screening may never reach.

In an AI-enabled hiring environment, organisations need more than surface confidence. They need clearer visibility.

That visibility still has to be interpreted properly. AI does not make the final call. Human analysts and responsible decision-makers do.

Accountability still lands with people

The original question from the session is the one to come back to.

When AI lies, who is accountable?

Inside the organisation, accountability does not disappear because a tool produced the wording.

At the point of recruitment, accountability does not disappear because an application looked polished, consistent, and professional.

In both cases, the responsibility sits with the people and processes making the decision.

That is why this conversation matters.

AI is making it easier to generate plausible output on every side of the hiring process. If your controls were designed for a world where poor information looked obviously poor, they are already under pressure.

Final thought

Applicants have always been able to lie.

What changes with AI is the polish, consistency, and speed with which some people can do it.

For leaders responsible for safeguarding, workforce vetting, compliance, and reputation, the question now is simple.

In an AI-assisted hiring environment, polished no longer means trustworthy.

The question for leaders is no longer whether an application looks credible.

It is - what in the process has actually been verified?

‍