AI-perpetrated sexual harassment demands that we learn in contextual ways
Experts in the field of violence against women and girls have been telling us for over 5 years about the risk, scale, and impact of deepfake imagery. But it’s taken a large-scale campaign of degradation and humiliation, perpetrated by users of X generative AI tool, Grok to bring the issue into clear focus. Yes, AI systems really can now be prompted to produce non-consensual, deepfake, pornographic material with ease, scale, and speed.
In the UK, lawmakers have responded by fast-tracking legislation that makes the creation or requesting of nonconsensual sexually explicit deepfakes a criminal offence, not just their distribution. This legal shift has raised an immediate question in HR forums in the past couple of weeks – is it their responsibility to educate staff on these changes?
When it comes to sexual harassment prevention, the answer is yes.
Under the Worker Protection (Amendment of Equality Act 2010) Act 2023, employers have a proactive duty to take reasonable steps to prevent sexual harassment. AI-enabled abuse and deepfake creation now sit squarely within that duty, because they are both workplace harassment risks and, increasingly, criminal conduct.
Policies assume understanding, but this concept is so abstract to many employees
To avoid risk, many organisations focus heavily on demonstrating compliance on paper. Updating policy wording or adding references to online behaviour have been immediate responses recommended in the HR community. On paper, this appears compliant. In practice, it is insufficient.
Policies assume understanding. They assume employees can identify what deepfakes are, recognise when behaviour crosses into criminality, and appreciate the harm caused by non-consensual sexual imagery. Evidence from training environments suggests otherwise.
Without explicit education, employees routinely misclassify AI-generated sexual content as inappropriate or distasteful rather than illegal. Some see it as novelty or technical curiosity. Some completely do not believe it is even here or happening. These gaps create space for harm to go unchallenged.
Scenario based learning can have the biggest impact
Over the past two years, I have measured the impact of different learning approaches on employees’ ability to recognise immoral and illegal sexual misconduct. Using paired t-tests to assess understanding before and after training, the findings have been consistent.
When learning relies on legal definitions and abstract concepts alone, improvements are limited. When learning is scenario-based, grounded in realistic workplace contexts, the results change significantly.
On average, there is a 20 percent increase in participants’ ability to correctly identify both immoral and criminal sexual misconduct when training uses contextual examples rather than conceptual definitions alone.
This is not coincidental.
What neuroscience tells us about recognising harm
The brain does not learn best through rules in isolation. Definitions are processed cognitively, but they are poorly retained under pressure. Scenarios activate additional neural networks linked to emotion, memory and social reasoning.
In simple terms, people remember situations far better than statutes.
Sexual harassment and exploitation are rarely experienced as clear-cut policy breaches in the moment. They emerge through ambiguity, social pressure and normalisation. Scenario-based learning allows the brain to rehearse recognition and response, making it more likely that learning translates into action.
This matters when considering AI-enabled harm, which often feels abstract until it happens to someone nearby.
Where should employers start
Education in this space must be explicit and contextual.
As a starting point, employees need contextual clarity on three things.
1. First, what deepfakes are. AI-generated images or videos that convincingly depict real people in sexualised contexts without consent.
2. Second, that creating or requesting sexually explicit deepfakes is now a specific criminal offence in the UK, not merely misconduct.
3. Third, why non-consensual sexual imagery causes harm, regardless of whether it is physically created, digitally generated or shared privately.
In this process it is important not to demonise the technology itself. AI tools do not invent sexual violence. They replicate and accelerate behaviours rooted in entitlement, power and control.
The same motivations that once drove physical acts of humiliation now operate through digital means. The tool has changed. The intent has not.
AI has huge potentials to accelerate many of our organisations, but we must strike a balance in managing both the risk and perceptions of it carefully, to engage all of our employees in our AI journey.