The Brief
- A new King’s College London report examines how existing inequalities could shape AI’s impact on women’s work. It offers a theoretical framework, rather than a new employment forecast.
- Exposure to AI describes tasks that technology could change. It does not establish how many workers will lose their jobs.
- Employers should examine access to useful tools and paid learning time before treating adoption as a measure of ambition.
- Clear standards for AI-assisted work should apply equally to everyone, including when managers allocate credit and promotions.
- A credible rollout should track job quality, recruitment and progression alongside productivity, and give workers a way to challenge unfair decisions.
Research finds that AI could deepen gender inequalities
An AI rollout can look successful while making a workplace less fair. Output rises, the software is widely used, and nobody immediately loses a job. Yet some employees gain authority over the new system while others spend their day checking its mistakes. One group gets development opportunities; another gets a higher target.
That is a scenario employers should test for when they introduce AI. A productivity figure alone cannot tell them who has benefited.
AI could deepen gender inequalities because its arrival changes the value assigned to work, the resources available to do it and the decisions made about the people performing it. A new report from King’s College London gives that conversation a useful structure.
What does the King’s report actually tell us?
When AI Met Gender: Gendered Impact of AI on the Labour Market, by Dr Hyojin Seo and Professor Heejung Chung, comes from King’s Global Institute for Women’s Leadership. It examines four connected areas: access to technology, organisational practices, occupational segregation and the undervaluing of women’s work, and insecure employment.
Its argument is that existing inequalities shape both the opportunities and risks AI creates.
Unpaid care can restrict learning time; workplace expectations can affect how AI use is judged; and workers with weaker employment security may have less protection during change. The report develops a theoretical framework using existing literature. It does not measure a new, nationwide AI employment penalty.
For Indian employers, that makes it a starting point for investigation. A UK discussion cannot supply an India-specific answer, but it can reveal questions missing from an organisation’s own rollout.
Who gets the tools before everyone is judged on the results?
At the report’s launch discussion, OECD economist Anna Milanez described how unequal access can arise: early tools go to technology teams or volunteers, premium access goes to managers, and part-time workers have less time to experiment. These allocation choices can favour groups in which men are more strongly represented.
Consider an illustrative Indian services company introducing a paid AI assistant. Its first licences go to team leaders. Everyone else can use a limited version, subject to approvals. Training takes place after the working day, and the company later asks managers to nominate its most innovative employees.
The process has already influenced who can demonstrate innovation. A person’s reluctance to experiment may reflect restricted access, unclear data rules or an unavailable training slot. None of those tells an employer how capable that person could become with proper support.
There is a straightforward alternative. Allocate tools according to tasks, publish the criteria and provide practice time within paid hours. Repeat sessions across shifts. Include part-time staff and people returning from leave. Give workers safe examples they can practise on without uploading customer information.
A course invitation should never be mistaken for an equal opportunity to learn. This is part of the wider problem examined in our coverage of barriers for women in tech: entry into a workplace does not settle access to progression.
Can women be criticised for using the very tools they are told to learn?
The report raises this tension, drawing on research about how AI-assisted work is evaluated.
One cited study, Code For Good Now’s The AI Judgment Penalty, tested responses from 1,000 UK adults to an identical AI-supported marketing CV bearing either a woman’s or a man’s name. It reports that respondents were twice as likely to question the female candidate’s competence.
That is a finding about judgements in a particular experiment. Participants were UK adults, not exclusively recruiters, and the design did not include a corresponding no-AI comparison. It therefore cannot establish how much additional bias AI disclosure created beyond existing gender bias. Nor does it measure Indian hiring outcomes.
Even with those limits, the managerial question is worth asking. Would the same use of AI be described as resourceful for one employee and evidence of dependence for another?
Employers should decide what acceptable assistance looks like before reviewing work. That includes disclosure, checking sources, protecting confidential information and taking responsibility for the final result. Evaluators should assess those behaviours consistently, using evidence of quality and judgement.
Our earlier examination of gendered rewards for AI assistants considers a different experiment. Together, these discussions remind us to inspect the human decisions surrounding technology, as well as the technology itself.
Does exposure to AI mean women’s jobs will disappear?
No. The International Labour Organisation’s March 2026 research brief, drawing on data from 84 countries, reports that 29% of female-dominated occupations are exposed to generative AI, compared with 16% of male-dominated occupations. Those percentages describe categories of occupations. They do not mean that 29% of women will lose their jobs.
The ILO expects changes to tasks, skills and working conditions to be more likely than widespread job losses for most occupations. Concentration in clerical, administrative and business-support work helps explain women’s greater exposure.
This is where employers need a more honest account of what a job contains. Preparing a document may be easy to automate. Knowing that the document contains an exception, contradicts a client commitment or needs an urgent escalation may require experience the task list barely records.
For an illustrative payroll team, generating a first response to a salary query is one task. Resolving an incorrect deduction before an employee’s rent payment is another. A faster draft does not make the surrounding responsibility disappear.
Before redesigning such a role, ask the people doing it to document exceptions, judgement calls and consequences of error. Then specify who will own those responsibilities after automation. Otherwise, a company can remove recognised expertise from a job description while continuing to depend on it informally.
What happens to the woman whose contract is simply not renewed?
The King’s report also directs attention to employment security: the effects of AI may differ between protected core staff and workers in less secure arrangements.
An Indian business applying that question should look beyond permanent headcount. Has spending on outsourced work changed? Are freelancers receiving fewer assignments? Have rates fallen while checking requirements increased? Has a junior vacancy disappeared without any corresponding training route being created?
These are proposed checks, not claims that the report measured those outcomes in India. They matter because a dashboard recording only redundancies can miss a substantial change in somebody’s livelihood.
The same principle applies to agency staff working alongside permanent employees. If their tasks change, the client and staffing provider should agree who supplies training, pays for learning time and handles a disputed performance assessment. The worker should not have to discover the gap between two organisations after something goes wrong.
Four myths that make the problem harder to see
Let us look at 4 myths that circulate in abundance on the internet.
Women need more confidence
Some may want support, but confidence is an incomplete diagnosis. A UK working paper by Fabian Stephany and Jedrzej Duszynski links adoption differences to perceptions of AI’s societal risks, including privacy and labour-market disruption.
It offers evidence that caution can have reasons beyond digital capability. [5] Ask what prevents use before prescribing a motivational workshop.
A computer makes hiring objective
Recruitment technology needs scrutiny. The UK Information Commissioner’s Office found fairness and privacy problems in its audits of recruitment-tool providers, including tools that allowed filtering on protected characteristics.
An employer should ask for evidence about the actual system and its proposed use, rather than accept a vendor’s general assurance.
Saving time guarantees a better job
Time saved could support learning, reduce overload or improve service. It could also become an automatic increase in targets. Employers should agree how gains will be used and examine quality, workload and pay after deployment.
AI must therefore be bad for women
That conclusion discards useful possibilities. Research published in The Quarterly Journal of Economics, covering 5,172 customer-support agents, found that access to a generative AI assistant increased issues resolved per hour by 15% on average, with substantial benefits for less experienced workers.
It studied one firm and does not establish a gender-equality effect. It does show that assistance can help people develop and perform.
What should an employer change before expanding its AI rollout?
The response should be concrete enough to appear in budgets, working arrangements and review meetings. CIC proposes the following checks, informed by the report’s questions and the supporting evidence.
| Decision | Practical change | Evidence to review |
| Access and learning | Fund task-appropriate tools and protected practice time across relevant roles and shifts. | Who receives access, attends training and can use the tool effectively. |
| Assessment | Establish consistent rules for permitted assistance, verification and credit. | Differences in ratings and progression, investigated within comparable roles. |
| Recruitment | Test the deployed system, explain its use and provide meaningful human review. | Errors, selection patterns, complaints and corrected decisions. |
| Job redesign | Map judgement and exception-handling before changing staffing or targets. | Quality, workload, earnings and career opportunities after deployment. |
| Worker involvement | Include affected employees and relevant contract workers in pilot reviews. | Issues raised, decisions changed and named owners for unresolved problems. |
HR, operations and technology teams should share responsibility for these checks. Aggregate results need careful interpretation: a difference is a reason to investigate, not automatic proof of discrimination. Monitoring should use lawful, proportionate data and protect people in small groups from being identified.
The UK regulator’s procurement guidance recommends checking bias mitigation, monitoring outputs and explaining AI use to candidates. These are useful operational principles for Indian businesses to adapt, rather than a statement that UK law applies to them.
A human reviewer also needs time, relevant information and authority to reverse a recommendation. Merely adding a person to an automated workflow does not make the review meaningful.
The Change in Content we need: A fairer share of the upgrade
The King’s report gives employers a reason to examine the organisation receiving the technology. Our response is to make fairness part of the business decision from the beginning.
Ask a woman to learn AI, and make the learning possible. Ask her to check its output, and recognise the expertise involved. Use a system to assess her, and give her a route to question the result. Redesign her role, and explain what happens to her earnings and future.
A company should be able to show who gained time, who gained skills and who gained opportunity from its investment. If it can only show faster output, it has answered a much smaller question.
Editorial note and sources
This analysis centres on the King’s theoretical report and checks selected claims against original research and institutional publications. Illustrative workplace situations and proposed interventions are CIC’s analysis, not reported case studies. No researchers or workers were interviewed. Findings from other countries are not presented as Indian prevalence estimates.
Sources
- Seo and Chung, When AI Met Gender: Gendered Impact of AI on the Labour Market. Principal report, especially the executive summary and framework sections.
- King’s College London launch coverage, 20 October 2026. Event observations are distinguished from the report itself.
- ILO, Gen AI, occupational segregation and gender equality in the world of work, 5 March 2026.
- Code For Good Now, The AI Judgment Penalty, 2026. Proprietary experimental research; design limitations are stated in the article.
- Stephany and Duszynski, Women Worry, Men Adopt, January 2026. Working paper, not presented as a settled causal explanation of all adoption differences.
- ICO, AI recruitment audit findings, 6 November 2024.
- Brynjolfsson, Li and Raymond, Generative AI at Work, QJE, 2025.
- ICO, Questions when procuring AI recruitment tools.
