The challenges and opportunities of AI in sustainability work
Climate work and supply chain management still involve a lot of manual effort: digging through spreadsheets, chasing data and comparing alternatives without knowing which numbers to trust. That isn’t just inefficient. It’s unnecessary. Technology can do that part faster, cheaper and better.
The point isn’t to replace people. It’s to let people focus on what they do best: judgement, priorities, relationships and action.
AI can be confidently wrong
In 2025, a Norwegian municipality made headlines after publishing an important report that contained AI-generated content. Much of the information was wrong, and several sources were incorrect or entirely fabricated. It’s a textbook example of AI being confidently incorrect, and a good reminder to treat AI output with healthy scepticism.
AI isn’t perfect, and won’t be any time soon. But used properly, its value can outweigh the occasional mistake. We accept that people make mistakes. Shouldn’t we accept the same of AI, as long as the right safeguards are in place?
Four ways to reduce the risk of AI errors
- Understand the limitations of AI models: know how a model is trained, what data it relies on and where its blind spots are.
- Write precise, well-structured prompts: clear input leads to more reliable output, while vague prompts make hallucinations and irrelevant results more likely.
- Use multi-agent or cross-checking systems: AI agents that verify or challenge each other’s output help surface inconsistencies and catch errors before they spread.
- Always cite and verify sources: never take AI-generated references at face value, and check that sources actually exist, especially in high-stakes work.
How we make AI-driven ESG risk assessments reliable
Reliability has been central to how we’ve built AI into supplier due diligence, now part of automated risk screening in our supplier management. AI helps monitor news about your suppliers, identify potential risks and assess suppliers against key environmental, social and governance (ESG) risk factors, supporting due diligence under the Norwegian Transparency Act (åpenhetsloven) and the UK Modern Slavery Act.
So how do we make sure AI makes reliable, responsible evaluations in such a sensitive context? With a network of specialised agents. Each agent has one focused task and builds on the work of the one before it, which gives both speed and quality control.
Feasibility agent
Before any assessment starts, the feasibility agent checks whether there is enough reliable information about a company. It asks:
- Is this a real, identifiable company?
- Can it be confidently linked to a specific sector and location?
- Is there enough online data for a meaningful assessment?
If the answer is no, the company is left out of the company-level screening, and we fall back on sector-level risk scores from trusted sources such as the ITUC and the ILO.
Company context agent
For companies that pass, the company context agent gathers the detail the next steps need:
- Corporate structure
- Products and services
- Known partnerships and affiliations
This context gives the risk agents accurate, structured input to work from.
Risk agents
Next, one risk agent runs for each ESG risk category, such as child labour, corruption or environmental violations. Each risk agent:
- Gathers relevant data for its risk domain, using a defined set of rules
- Analyses company-specific patterns, red flags and context
- Scores the risk from 1 to 5, using a ruleset that weighs both the likelihood and the consequence of that risk
Because each risk is assessed on its own, every score is transparent and specific to its domain. Running the agents in parallel keeps the assessment both detailed and efficient.
Verification agent
Once the risk agents are done, the verification agent runs a final quality check across the whole network. It reviews:
- The consistency and plausibility of each risk score
- Whether each assessment is backed by enough relevant evidence
- Any anomalies, contradictions or missing context
If it finds a problem, it flags the result for review or reprocessing. That keeps the final ESG risk profile coherent, reliable and traceable, and reduces the chance of errors slipping through.
With the right safeguards, AI accelerates sustainability performance
Specialised agents working together, backed by a robust verification step, significantly reduce the risk of AI errors. Vigilance still matters, so we keep refining both the technology and the processes around it. That way, you get the speed of AI with decisions based on accurate, reliable and ethical data, and can navigate sustainability with greater confidence.
