Kenya has released a draft policy that could fundamentally alter how global technology companies treat the workers who train, test, and moderate the AI systems embedded in everyday software—from Microsoft Copilot to enterprise cloud services. The proposal, published on July 25 by the Ministry of Information, Communications and the Digital Economy, introduces a fair-pay reference framework benchmarked to international rates and mandates mental health safeguards for content moderators and data annotators, addressing the often overlooked human layer behind generative AI.

Concrete Changes: What the Draft Policy Puts on the Table

The draft "Kenya Artificial Intelligence and Other Emerging Technologies Policy" is not yet law, but it outlines the government’s explicit intent to set occupational standards for AI-related work. Key provisions include:

  • Fair-pay reference framework: Employers would be required to benchmark compensation against international rates for equivalent roles, with mandatory public disclosure of how their pay compares.
  • Written contracts: All workers must receive understandable employment agreements.
  • Psychosocial support: Mandatory mental health services, including confidential counseling and trauma-informed care.
  • Harmful-content safeguards: Limits on exposure to disturbing material, with rights to rotation and reassignment without penalty.
  • Grievance mechanisms: Accessible and safe reporting channels for workers to raise concerns.
  • Proportionate surveillance: Restrictions on workplace monitoring, requiring transparency and purpose limitation.
  • Duty of care: Employers—including international firms operating via outsourcing partners—must actively protect workers from occupational harms.
  • Compliance reporting: Regular public reporting on adherence to these standards.

The policy applies to any company employing data annotators, content moderators, or AI quality evaluators in Kenya, whether directly or through third-party vendors. Crucially, the policy doesn’t just target direct employers. It seeks to hold international technology companies responsible when they contract with local outsourcing firms, closing a longstanding accountability loophole.

The Workers You Never See: Annotation, Moderation, and Evaluation

Most Windows users never see the thousands of people behind the polish of their AI assistants. But every time Copilot summarizes an email or ChatGPT answers a prompt, a human has likely played a role in making that interaction safe and accurate. These workers fall into three main categories:

  • Data annotators label images, text, audio, or video so AI models can recognize patterns. They might outline objects in a photo, transcribe speech, or rank chatbot responses.
  • Content moderators review user-generated material—from social media posts to uploaded documents—flagging and removing harmful content. They routinely encounter violence, abuse, and other trauma-inducing material.
  • AI quality evaluators test model outputs for accuracy, bias, and safety, often comparing multiple responses to determine the best one.

Without these human judgments, generative AI could not function reliably. Yet the mental and financial well-being of these workers has often been an afterthought. Workers contracted through the outsourcing firm Sama, which handled projects for OpenAI and Meta, have reported severe psychological distress, low wages, and abrupt job losses—allegations the company has disputed.

Why This Matters: From Your Desktop to Boardroom Ethics

For home users, the draft policy could improve the consistency and safety of AI features like Windows Copilot by ensuring the human evaluators behind them aren’t burned out or checking out. When you dictate an email to Copilot, human evaluators have likely reviewed similar outputs to ensure the AI responds appropriately. Their mental state directly impacts the safety features you trust.

For IT professionals and admins, rolling out Microsoft 365 Copilot across an organization isn’t just buying a license; it’s wading into a supply chain that includes human intelligence. The Kenyan proposal provides a framework to question how vendors manage that chain.

For enterprise buyers, large-scale procurement departments should add AI labor standards to their vendor assessments. Just as you ask about data center sustainability or conflict minerals, you can now ask: “Are the data annotators and moderators who support this product guaranteed fair pay, mental health support, and safe working conditions?” The draft policy in Kenya suggests that such questions will become commonplace, and vendors who can’t answer them may face reputational or regulatory risk.

How Kenya Became the World’s AI Back Office

Kenya’s rise as an outsourcing hub wasn’t accidental. The country offers a large, English-speaking workforce, a growing tech sector, and competitive costs. Over the past decade, major outsourcing firms like Sama have set up large moderation centers in Nairobi and other cities, capitalizing on the global AI boom. But the rapid growth also drew scrutiny. Reports by investigative journalists and human rights groups highlighted worker grievances—trauma from graphic content, unpredictable pay, and a lack of support.

The government’s draft policy is its first comprehensive response. It explicitly acknowledges that existing labor laws don’t adequately address the unique risks of AI work, from psychological harm to opaque surveillance. By proposing a regulatory framework, Kenya is signaling its intent to move beyond a low-cost labor model toward a more sustainable, rights-respecting AI ecosystem.

Action Plan: What You Should Do Now

The draft policy is a signal, but you don’t have to wait for it to become law. Here are steps you can take today, depending on your role:

  • For consumers and power users: Stay informed. If you care about ethical tech, support companies that disclose their AI labor practices. Use your voice on social media or feedback channels to ask for transparency.
  • For small business owners and power users selecting AI tools: Look for public statements on responsible AI that include supply chain labor. Your choices signal demand for ethical practices.
  • For IT admins and enterprise buyers: Add the following questions to your vendor assessments:
    • Do you have a human rights policy covering AI supply chain workers?
    • Do you require outsourcing partners to provide written contracts, fair pay, and mental health support?
    • How do you audit and enforce these requirements?
      Monitor regulatory developments; Kenya might influence other jurisdictions, and being proactive now can give you a competitive edge.
  • For developers: If you’re using AI APIs, consider the full stack. Push your employer or platform to adopt ethical sourcing.

Outlook: A Draft That Could Redraw the Map

The public consultation period invites feedback from industry and workers. The final policy, expected later this year, will need strong enforcement to be meaningful—a designated authority, inspection powers, and real penalties for non-compliance. Without that, companies might simply pay lip service while continuing practices as usual.

But the global trend is toward greater accountability. The EU’s Digital Services Act and proposed AI Liability Directive, along with worker protection initiatives in the U.S., suggest that Kenya’s move is not an outlier. For Microsoft users, the bottom line is that the AI you use is not fully automated. It’s a human-machine partnership, and Kenya’s draft policy is a critical step toward ensuring that partnership is fair. Keep an eye on how the world’s biggest tech companies respond—their actions will reveal how seriously they take the people behind the products.