What happens when a hiring manager asks a chatbot about you�and it confidently invents a baby you don�t have? That�s what Financial Times journalist Olivia Gagan discovered when Grok, xAI�s assistant, falsely stated she was on maternity leave? It�s a jarring anecdote, but also a timely warning: as recruiters lean on generative AI (software that creates text, audio, or video from prompts), misinformation about candidates can quietly seep into decisions?
AI is already in the funnel�often informally
According to Boston Consulting Group, 70% of companies experimenting with generative AI are using it in HR, with talent acquisition the top use case, the FT reports? Recruiters say they sometimes consult general-purpose chatbots to �get a read� on a candidate�despite the risk of fabricated details and inferred personal attributes? Gartner�s Jamie Kohn calls it �quite concerning,� noting that AI can be wrong and even reveal protected information that shouldn�t influence hiring?
That mirrors the field experience captured by the BBC: with vacancies down 12% year over year in the UK and applications per role up 65%, some employers are turning to automated phone or video interviews to cope? Cera, a care provider, told the BBC that its AI phone screener saves two recruiter-days a week and cuts screening costs by two-thirds? The downside? Glitches and shallow filters? A TestGorilla AI video interview malfunction affected a small number of candidates, while jobseekers say bots often miss the �bigger picture?�
The �invisible interview� meets AI hallucination
Generative systems also turbocharge a parallel trend: careers are increasingly shaped by digital signals candidates don�t fully control? As ZDNET puts it, we�re in an �Invisible Interview� era, where your public footprint, agent-mediated interactions, and platform profiles speak before you do? Gartner estimates that by 2027, 75% of hiring processes will include AI proficiency testing; paradoxically, through 2026, 50% of organizations may add �AI-free� skills assessments to counteract critical thinking atrophy? For candidates, that means mastering tools without letting them define you?
Scale matters here? Comscore data, cited by ZDNET, shows Grok is among the fastest-growing assistants�472% year-to-date growth on mobile�while Google�s Gemini surged 971% on desktop? ChatGPT remains dominant overall? If recruiters are even informally consulting these systems, errors can multiply quickly�and quietly?
Legal exposure without clear guardrails
The FT highlights a real compliance trap: if AI falsely infers or exposes protected characteristics (like pregnancy) and that shapes a decision, companies could face discrimination claims�whether or not the information is true? UK regulators have pushed the onus onto employers to use AI �at their own risk,� conduct impact assessments, and be transparent with candidates? Korn Ferry�s Bryan Ackermann calls generative AI in sourcing �as much an opportunity as it is a risk,� urging strong governance to match the efficiency gains?
US policy could centralize�or complicate�rules overnight
In the US, rules may shift dramatically? President Donald Trump says he will sign a �ONE RULE� executive order to block state AI regulations and impose a single federal standard, according to TechCrunch and the FT? Proponents argue a patchwork of state rules is unworkable; critics, including Republican figures, warn it undermines state authority to protect residents from harms like deepfakes and other AI abuses? While the draft order targets broad AI governance�citing laws from California to Tennessee�any sweeping preemption could spill into employment tech, altering how bias audits, transparency, and automated decision tools are governed across states?
Whether centralization clarifies compliance or invites legal limbo will determine how HR adopts AI in 2025? For now, HR leaders should assume scrutiny will tighten, not loosen?
What employers should do this quarter
- Ban off-the-shelf chatbots for candidate research? Use vetted tools with auditable logs, and document intent and use-case boundaries?
- Institute a �do-not-infer� policy? Prohibit models from inferring protected traits; monitor prompts and outputs for violations?
- Require adverse-impact testing and error reporting? Treat AI misstatements as incidents with remediation and candidate notification?
- Provide a right-to-correct? Offer candidates a clear channel to dispute automated outputs and ensure corrections propagate across systems?
- Balance automation with depth? Use structured �AI-free� work samples to counterbalance automated screens and reduce false negatives?
For candidates, curating accurate public profiles and proactively correcting AI summaries can reduce risk? But as Gagan�s experience shows, managing the �third identity��the version of you AI describes�may soon be as important as polishing your resume?
The bottom line: automation is rushing into hiring because volume pressures demand it? The winners will be those who capture efficiency without outsourcing judgment�and who can answer a simple question with evidence: when your AI says a candidate just had a baby, how do you know it�s true?