Can AI align with your sustainability and ethically driven goals?

AI usage has exploded over recent years, increasingly reshaping developer-focused roles — and seemingly coming full circle, with some organisations rediscovering that hiring juniors is cheaper than burning AI tokens on simple tasks. That trade-off alone is something I could talk about for hours, but I’ll just say this: every senior developer was once a junior. Stop hiring juniors, and you’ll run out of seniors soon enough.

That’s not really what this post is about, though. It’s about the quieter question sitting underneath most organisations’ AI rollouts: what does using this technology responsibly actually mean, in practice, day to day?

“Alignment” in AI research usually means something narrow and technical: does a model behave the way its creators intended? But for the organisations adopting these tools — not building them — alignment needs to mean something broader. It’s alignment with the goals, values, and commitments the organisation already claims to hold. Here’s what I think that has to cover.

This is something I’ve recently added to the webinars I’ve been delivering through Aline after AI being the topic of may of the questions afterwards, but it definitely needs expanding on so here are my own thoughts.

Alignment with the SDGs

Most organisations of any size now report against some version of the UN Sustainable Development Goals, or at least gesture at them on a sustainability page nobody reads. AI adoption rarely gets held to the same standard. Procurement decisions — which model, which vendor, how it’s deployed — happen in a completely separate conversation from ESG commitments, as if the two are unrelated.

They’re not. Decent work and economic growth (SDG 8), responsible consumption and production (SDG 12), and climate action (SDG 13) are all directly implicated by how an organisation adopts AI, not just whether it does. If your organisation has made public commitments in these areas, your AI vendor’s labour practices and energy sourcing are quietly part of that commitment, whether anyone’s auditing it that way or not.

Digital ethics

Digital ethics in an AI context usually gets discussed as a model-output problem: bias, hallucination, toxic content. Those matter, but they’re the visible tip of it. The less visible part is upstream — who trained the model, under what conditions, and what the organisation buying access to it actually knows about that.

Most companies deploying AI tools today have little idea what’s in their vendor’s supply chain, and most vendors don’t make it easy to find out. Ethics-as-a-checkbox (a policy document nobody reads) is not the same thing as ethics-as-a-practice (asking your vendor a hard question, and being willing to switch if you don’t like the answer).

Literacy

The most underrated organisational risk right now isn’t AI misuse — it’s AI illiteracy. Not knowing what a model can and can’t reliably do. Treating a confident answer as a correct one. Losing the internal capability to check its work because the juniors who’d have caught the error got automated out of the pipeline before they built that judgement.

Literacy has to sit above the technical team too. The people making adoption and procurement decisions need enough grounding to ask good questions — not just enough enthusiasm to sign the contract.

Frivolous use and the rise in energy demand

This is the one closest to my own background. Global data centre electricity demand grew 17% in 2025, and consumption from AI-focused data centres specifically surged by roughly 50% over the same year — a trajectory the International Energy Agency expects to roughly double overall data centre demand between 2025 and 2030 (IEA). None of that is inherently a problem; useful compute has a cost, same as anything else. But a meaningful share of that demand goes toward low-value, low-effort use: asking a model to do in ten prompts what a five-minute search would’ve answered, generating throwaway images for a Slack message, running an agentic workflow on a problem a spreadsheet already handled.

Organisations serious about their own sustainability commitments have to treat frivolous AI use as a real line item, not a rounding error — the same way they’d treat unnecessary business travel or idle servers.

So can AI help with sustainabiltiy and ethics?

It can, but be wary

AI can absolutely help organisations move faster on everything above — sustainability reporting, accessibility, literacy, even reducing energy use elsewhere in the business. But “can” is doing a lot of work in that sentence, and it depends heavily on whose model you’re using and what they actually stand for.

Two examples make the point better than an abstract argument does. A 2023 TIME investigation found that OpenAI had used outsourced Kenyan workers, paid roughly $2 an hour, to label graphic and traumatic content so ChatGPT could be trained to filter it out — work some of those workers later said left them with lasting psychological harm (TIME). That’s not a hypothetical harm buried in a training run; it’s a labour and ethics failure with a name and a paper trail.

Contrast that with Anthropic’s decision last year to give Claude the ability to end conversations with persistently abusive users, framed explicitly as an experiment in “model welfare” — protecting the system, rather than the person on the other end of it (Anthropic). Reasonable people can disagree about whether that’s the right mechanism, or even a sensible one. But it’s a fundamentally different posture toward the ethics of building AI: one treats the pipeline behind the model as an externality; the other treats it as part of the product.

That’s the wariness organisations need. Alignment isn’t something you get by picking any AI vendor and layering a usage policy on top. It’s a property of the whole chain — labour, energy, literacy, and intent — and the model an organisation chooses says more about its actual values than the policy document sitting next to it.