The Employment Rights Act 2025 wasn’t written with AI specifically in mind. But several of its genuinely new protections land directly on decisions that AI systems now routinely influence, performance monitoring, dismissal, enforcement. This piece looks at three real points where the two collide, not a broad speculative survey, but the specific places where a new legal right and an AI-driven decision actually meet.

This connects two threads already covered separately in this series: the Act’s own changes, and the mechanics of how AI systems produce decisions that carry legal weight. Here’s where they intersect in practice.

Explore: Employment Law Changes 2026: HR Guidance on the Employment Rights Act 2025, for the full picture of the Act.

Explore also: AI Grievances and Employment Tribunals: A Guide for HR Teams, for the AI-specific legal framework this piece draws on.

The Six-Month Qualifying Period Meets AI Performance Scoring

From 1 January 2027, unfair dismissal protection kicks in after six months’ service instead of two years, a change already covered in detail elsewhere in this series. What that means for AI specifically is worth spelling out.

A significant, current legal development sharpens this collision point. Section 80 of the Data (Use and Access) Act 2025 replaced UK GDPR’s Article 22 with four new articles, 22A to 22D, in force since 5 February 2026. The old regime prohibited solely automated decisions with significant effects by default. The new one permits them, provided specific safeguards are met:

• Information about the decision
• The ability to make representations
• A right to human intervention
• The ability to contest it

One detail is still unsettled: Article 22D allows further regulations defining exactly what counts as “meaningful” human involvement, and none had been published as of writing. The core safeguards are law now; some of the finer interpretive detail is still to come.

An AI system flagging someone for dismissal within their first six months is exactly the kind of “significant decision” this regime is built for. Fox Williams puts the practical consequence plainly: employment tribunals are likely to examine closely whether it was reasonable for an employer to rely on AI-generated data when deciding to dismiss someone, particularly in a case where the old two-year buffer no longer applies.

The two changes compound each other. More dismissals now fall within scope of unfair dismissal protection at an earlier point, and any of those dismissals influenced by an AI system now also has to clear a specific statutory safeguards test. Skipping either isn’t a paperwork gap, it’s exposure on two fronts at once.

The Fair Work Agency’s Own Possible Use of AI in Enforcement

The collision here runs in a different direction: not AI making decisions employers need to defend, but AI potentially being used against employers by the regulator itself.

The Fair Work Agency, established 7 April 2026 as an executive agency of the Department for Business and Trade, has real teeth: the power to inspect workplaces, compel production of payroll and contract records, issue underpayment notices with penalties, and, per Personnel Today’s reporting, bring employment tribunal claims on behalf of workers.

What it doesn’t have, at least not yet, is a large budget to match those powers. Personnel Today, in an analysis by Addleshaw Goddard employment partner David Palmer, puts the FWA’s 2026-27 funding at £60.1 million, a fraction of the Health and Safety Executive’s budget of over £350 million for a comparably broad remit. Palmer’s point is direct: given the FWA can already access a business’s payroll and working time data, and given how thin that budget is relative to its task, using AI to analyse large datasets for compliance issues is a plausible next step, not a confirmed one, but a genuinely credible one given the resourcing gap.

The practical implication for employers is straightforward. Whether or not the FWA actually adopts AI-assisted analysis, the underlying exposure already exists: payroll and working time records are exactly the kind of data an automated system could process quickly, and errors that might once have gone unnoticed in a manual spot-check become easier to surface at scale.

Uncapped Compensation Meets Algorithmic Dismissal

The third collision point ties directly to the single biggest financial change in the whole Act. From the same date the qualifying period shortens, 1 January 2027, the cap on unfair dismissal compensation disappears entirely, a change covered in full elsewhere in this series.

Combine that with the Articles 22A-22D safeguards from Section 2, and the stakes of getting an AI-influenced dismissal wrong change materially. A dismissal that skips the required safeguards, no real information given, no opportunity to make representations, no genuine human intervention, was already a compliance failure. From 2027, it’s a compliance failure with no ceiling on what it could cost if it results in a successful unfair dismissal claim.

One point is worth repeating plainly, because it’s the instinct most employers reach for first: relying on a vendor’s AI tool doesn’t reduce this exposure. As covered elsewhere in this series, a tribunal isn’t going to accept “the algorithm made the recommendation” as a defence, any more than it would accept “a manager made the call” as a defence to a manager’s decision. The employer applying the decision is the one the law looks at, regardless of what produced the recommendation behind it.

What This Means for Employers Using AI in Employment Decisions

None of the three collision points above require waiting for a claim, an inspection, or January 2027 to start addressing.

• Audit AI-influenced decisions specifically within the first six months of employment. These now carry both the shortened qualifying period risk and the Articles 22A-22D safeguards requirement at the same time.
• Build the four required safeguards into the process itself, not as an afterthought. Information about the decision, a genuine opportunity to respond, real human intervention, and a clear route to contest the outcome all need to exist before a decision is finalised, not reconstructed after a claim is lodged.
• Assume payroll and working time data could face automated scrutiny, not just a manual check. Whether from the Fair Work Agency or an internal review, the same records worth getting right for compliance reasons are exactly the kind of data an AI system can process quickly.
• Treat vendor-supplied AI tools as part of the employer’s own compliance responsibility, not a transfer of it. The uncapped exposure from 2027 applies regardless of who built the system making the recommendation.

How Avado Can Help

Understanding where a new legal right and an AI-influenced decision actually collide is exactly the kind of judgement HR needs to bring to decisions a system helps make. Avado’s HR Compliance for Managers course, presented by employment law specialist Amanda Chadwick, builds that judgement directly, covering the Employment Rights Act 2025 changes alongside the AI-specific legal framework covered throughout this series.

Explore HR Compliance for Managers and make sure every AI-influenced decision can withstand scrutiny from both directions!