The UK duty to prevent sexual harassment, and where AI fits

Employers in Great Britain have a legal duty to prevent sexual harassment of their staff, and from 30 October 2026 that duty became stricter. Work now happens in chat apps, video calls and shared drives, and generative AI makes it easy to produce sexual images or messages about a colleague. Employers are also offered AI tools that scan messages or take anonymous reports. This page explains the duty, the changes, and what it means when AI is either part of the problem or part of the employer's response. It covers England, Wales and Scotland.

The short answer: an employer must take all reasonable steps to prevent sexual harassment of its employees in the course of their employment (Equality Act 2010, section 40A). Since 30 October 2026 the word "all" is in the duty, and employers are also liable when a third party, such as a client or customer, harasses an employee and the employer failed to take all reasonable steps to stop it. Where a tribunal upholds a sexual harassment claim and finds the duty was breached, it can raise compensation by up to 25%.

What the law says

Harassment includes unwanted conduct of a sexual nature that has the purpose or effect of violating a person's dignity or creating an intimidating, hostile, degrading, humiliating or offensive environment for them (Equality Act 2010, s26(2)). In deciding the effect, a tribunal must take into account the person's perception, the other circumstances and whether it is reasonable for the conduct to have that effect (s26(4)). Conduct does not have to happen in person to meet this test.

Section 40A says an employer must take all reasonable steps to prevent sexual harassment of its employees in the course of their employment (s40A). The duty first took effect on 26 October 2024 as "reasonable steps"; the word "all" was added from 30 October 2026 by the Employment Rights Act 2025 (SI 2026/1065, reg 3). An employee cannot bring a claim on section 40A alone. It is enforced by the Commission for Equality and Human Rights, and through a compensation uplift in a tribunal claim (s40A(3)).

Harassment by clients, customers and other third parties

From 30 October 2026, an employer must not permit a third party to harass an employee. It permits harassment only if the harassment happens in the course of the employee's employment and the employer failed to take all reasonable steps to prevent it (s40(1A) to (1C)). This applies to harassment related to any of the protected characteristics listed in section 26, not just sexual harassment. For staff who deal with the public through chat, email or video, it covers abuse received through those channels in the course of the job.

Liability for colleagues, and the reasonable steps defence

Anything an employee does in the course of employment is treated as also done by the employer, whether or not the employer knew or approved. The employer has a defence if it shows it took all reasonable steps to prevent the employee from doing that thing or anything of that description (s109). So if one employee uses an AI tool to create and share sexual images or messages about a colleague through work systems, the employer may be liable unless it can show it had taken all reasonable steps beforehand.

The compensation uplift

Where a tribunal finds sexual harassment of an employee and orders compensation, it must consider whether the employer also breached section 40A. If so, it may order an uplift reflecting the extent of the breach, of no more than 25% of the compensation (s124A).

Regulations on reasonable steps

Since 30 October 2026, section 40B lets the government specify by regulations which steps count as reasonable. These may include assessments, published plans or policies, and steps on reporting harassment and handling complaints (s40B). Check whether any regulations have been made before relying on a list of steps.

Where AI fits

• AI as a source of harm: image and text generators can be used to create sexual content about real colleagues. Workplace policies on acceptable use of AI tools and on conduct should say plainly that this is harassment and will be treated as misconduct. See writing an AI policy for employees.
• AI in prevention and reporting: tools that flag messages or take anonymous reports may help, but they are not a substitute for a risk assessment, training, a clear complaints route and people who act on reports. A tool that misses obvious abuse, or alerts nobody, does not show that all reasonable steps were taken.
• Monitoring and data: scanning staff messages is processing of personal data and must have a lawful basis and be proportionate. See AI monitoring at work.
• Customer facing work: where staff handle chats or calls with the public, think about how abusive users are blocked, how staff can end a conversation and how incidents are recorded, given the third party rule.

Reporting and confidentiality

Since 6 April 2026, a disclosure that a worker reasonably believes is in the public interest and tends to show that sexual harassment has occurred or is likely to occur can be a qualifying disclosure under whistleblowing law (Employment Rights Act 1996, s43B(1)(da)); see UK whistleblowing law and AI. The 2025 Act also inserted section 202A, which will make terms in agreements void so far as they stop a worker alleging or disclosing harassment or discrimination. On 31 October 2026 legislation.gov.uk showed it commenced only for specified purposes, so check its status before relying on it (ERA 1996, s202A).

For harassment related to race, religion, disability, age and the other listed characteristics, see AI and harassment related to protected characteristics. See also UK equality law and AI and grievances and discipline when software is involved. This page is general information, not legal advice.