An AI pipeline read the public disclosures of five semiconductor companies and found the same risk dominating all of them: trade restrictions. That is not an analyst guessing. It is a machine reading annual reports, 10-Ks, sustainability filings and investor decks, then ranking 76,207 separate risk and opportunity items against expert judgment.

Here is what surprised me. This is not a demo. An independent check found 92.6% of the extracted items valid, and the rankings matched human experts closely on the things that matter most. In other words, the grunt work your risk team does by hand every quarter, reading supplier documents and hoping they catch the real exposure, can now run continuously and at scale.

Why This Paper Matters

Every executive with a complex supply base sits on a slow warning system. Risks to your suppliers are sitting in their own public documents, hidden in dense risk-factor paragraphs and geopolitical footnotes. Mostly nobody reads them until something breaks.

The 2020 to 2023 chip shortage halted car plants around the world and wiped out hundreds of billions in GDP. Governments pumped in money, Europe committed up to 43 billion euros through the Chips Act, America committed 52.7 billion dollars. And still the machines catch exposure late. The paper, built for the ALLPROS alliance under the European Chips Act, shows the missing piece was never capital. It was intelligence that refreshed itself.

Well… most companies run risk registers that go stale between quarterly reviews. This work demonstrates an AI Assessment for companies that turns supplier disclosures into a live, ranked matrix your Chief AI Officer and operations team can actually act on, then re-run whenever new filings drop.

Methodology, Explained Simply

Think of it as a librarian with superhuman reading speed. The system pulls the public corporate documents of each company, then uses large language models to read every chunk and pull out the risks and opportunities the company itself describes. It is not inventing anything. It is structuring what suppliers already disclosed.

The hard part is noise. Five companies produced over 166,000 raw extractions, and most say the same thing in different words. So the pipeline clusters near-identical items, merges them into one, and links everything into a knowledge graph that records where each risk came from and what it relates to. That provenance matters, because when a risk moves up your list you need to know exactly which filing flagged it.

Then comes the part I would watch closely at any vendor. Ranking uses three layers: an algorithmic formula, an LLM relevance adjustment, and human expert validation. It is automation with a human review loop built in, not a black box. That is precisely the design your own AI risk system should copy, because it is how you get both speed and trust.

The whole thing read Intel, Infineon, Texas Instruments, Air Liquide and Siltronic across the value chain. What emerged was a shared warning and a shared opportunity. Trade restrictions and tariffs dominate every company. Meanwhile the biggest upside, electrification and market growth, largely does not overlap between companies, which is a rare chance for an alliance to grow together instead of competing for the same chips.

Results and Practical Insights

Start with the number that earns trust: 92.6% of the 76,207 scored items were independently judged valid, and rankings agreed with experts at an average Spearman correlation of 0.55 for risks and 0.72 for opportunities. The agreement on opportunities was notably tighter, which makes sense, opportunities are less ambiguous to spot than the best of many bad risks.

Trade restrictions were the dominant systemic risk across all five companies, a board-level signal that tariffs are not one company’s problem but the whole industry’s shared exposure. The model caught the nuances too. For Infineon, algorithmic ranking alone hit a 0.97 correlation with the expert, while a manufacturing-concentration risk the human expert cared about vaulted from rank 77 to number one once relevance was weighted in. That single jump is the whole argument for keeping a person in the loop.

Year over year the picture moved too. Revenue volatility overtook regulatory compliance as the top risk in 2025, and competitive pressure jumped from rank 15 to rank 8. Risks are not static; they are not a one-time assessment, and a system that only refreshes quarterly will keep telling you about last year’s problem while the current one silently climbs the list.

Key Takeaways for Operations Leaders and Your Chief AI Officer

Thanks to All Authors