OSLO, Feb 26 - Norway's sovereign wealth fund, the world's largest at $2.2 trillion, is applying artificial intelligence to its investment screening process to detect potential environmental, social and governance (ESG) and integrity concerns. The fund holds stakes in about 7,200 companies globally and owns roughly 1.5% of all listed stocks, making early identification of risk material to its mandate.
The fund's investments are benchmarked to an index set by the finance ministry, with equities tracked against the FTSE Global All Cap index. Each time that benchmark adds new companies, the fund's manager, Norges Bank Investment Management (NBIM), must screen those firms before they are included in the portfolio.
Since 2025, NBIM has employed large language models to screen every company on the day it enters the equity portfolio, scanning public information sources that traditional data vendors often omit. According to NBIM's annual responsible investment report, the AI-driven system acts quickly: "Within 24 hours of our investment, the AI tools flag new companies in the fund’s equity portfolio with potential links to, for example, forced labour, corruption or fraud."
The manager says the approach has had practical effects on portfolio composition. "In multiple instances, we identified and sold these investments before the broader market reacted to the risks, avoiding potential losses," NBIM reported, indicating the AI screening has supported timely divestments when red flags emerged.
NBIM also notes the particular utility of these tools for small companies in emerging markets, where data coverage by vendors can be sparse and international media attention limited. The fund explains that relevant reporting may appear only in local outlets and languages, and that controversies which might point to systemic failures in risk management can go unreported in global media. "News may be limited to small media outlets in local languages, and controversies suggesting systemic failures in risk management may go unreported in international media," the report said.
The use of large language models is described as a complement to existing screening procedures rather than a replacement for them. The rapidity of the AI scans aims to surface publicly available information that conventional vendor feeds might miss, enabling the fund to act early when integrity or governance concerns arise.
Context and mechanics
The fund's approach ties screening directly to benchmark composition - every time the FTSE Global All Cap index introduces new constituents, NBIM applies its AI process on day one of exposure. The manager emphasises that these tools help flag potential connections to forced labour, corruption or fraud, and that early detection has, in several cases, led to divestments ahead of wider market recognition.
Limits in coverage
NBIM highlights data coverage gaps as a driver of the technology's value. Smaller listed companies, particularly in emerging markets, may not be well represented by commercial data vendors and may only be covered in local media. The AI models are used to scan such public sources in order to reveal issues that otherwise might remain obscure.