companyMost of a 10-K risk factor section is boilerplate, and reading one on its own tells you little. Put it next to last year's version and it gets a lot more useful, because the differences show what management started worrying about in writing, with legal liability attached.
We compared Item 1A in the last two 10-Ks of 14 large U.S. companies: Alphabet, Amazon, Boeing, Eli Lilly, ExxonMobil, Intel, JPMorgan Chase, Meta, Microsoft, Nvidia, Pfizer, Tesla, UnitedHealth and Walmart. For most of them that means the FY2024 filing from early 2025 against the FY2025 filing from early 2026. We scored every sentence for sentiment and matched it against the prior year, so each one is tagged as new, edited, carried over or deleted.
The short version: "ESG" went from 19 mentions to none, though most of the underlying risks are still there under other names. Microsoft, Nvidia and Alphabet all added language about AI capacity, financing and counterparties. Tariff mentions rose 66% and no company cut them. AI mentions rose 44% overall, and Nvidia was the one company where they fell. And the tone of the new sentences told us more than the tone of the whole section. Where a risk section got more negative, the added sentences were roughly 1.5x to 6.5x more negative than the ones that were deleted.
1. ESG got renamed
In the prior-year filings, these 14 risk sections used "ESG" 19 times. In the latest filings they use it zero times. Walmart had the most mentions (8), followed by Tesla (4), JPMorgan (3), Intel (2) and Boeing (2), and all five dropped the term.
Most of the risks are still there with new labels:
| Company | Prior-year wording | Latest wording |
| Walmart | stakeholder expectations with respect to our ESG efforts | stakeholder expectations with respect to our social and environmental efforts |
| Nvidia | increased scrutiny from shareholders, regulators and others regarding our corporate sustainability practices | scrutiny from shareholders, regulators and others regarding our corporate sustainability practices |
| ExxonMobil | safety, business controls, and environmental risk management | safety, business controls, and risk management |
| Pfizer | responsible business growth | responsible business practices |
| Tesla | standalone risk on stakeholder scrutiny of its ESG practices | removed |
Climate language thinned out as well. Mentions of "climate" fell 28%, from 78 to 56, with the biggest cuts at JPMorgan (16 to 5), Intel (8 to 2) and Walmart (8 to 4).
If you screen filings for "ESG" or "climate," you'll now undercount how much of this risk companies still disclose. Walmart, Nvidia, Exxon and Pfizer kept the risk and changed a few words.
2. AI infrastructure got its own risk factors
Microsoft, Nvidia and Alphabet each added risks about building and paying for AI capacity. None of these new risks were in their prior-year filings.
Microsoft broke one cloud and AI risk into several, covering infrastructure buildout, outages and capacity constraints, and supply. It now warns that if it overestimates demand, infrastructure could sit underused and lead to impairments. It also says tighter credit or higher rates could limit its ability to execute its infrastructure strategy. And it says it is securing energy resources for AI; mentions of power or electricity went from 4 to 11.
Nvidia added a new heading: "commercial arrangements expose us to counterparty risks." It discloses long-term capacity purchase obligations and financial guarantees, and says it has been asked to help finance customers' and partners' datacenter buildouts. It adds that it hasn't entered into any financing arrangements. Its supply risk heading also changed from "component availability" to "capacity availability."
Alphabet added a heading about disruptions to its ability to access future financing or manage its indebtedness. New text also covers large long-duration commercial agreements, excess capacity it might not be able to redeploy, and counterparties that might not pay.
Investors have spent the past year arguing about whether the AI buildout is overbuilt and how much of it runs on credit, including deals where suppliers help fund their own customers. Three of the biggest companies in that buildout have now put those same concerns in their 10-Ks.
3. Tariffs went from possible to actual
Tariff mentions rose 66%, from 56 to 93. Nine companies increased them and none reduced them.
| Company | Tariff mentions, prior year | Latest year |
| Tesla | 1 | 10 |
| Eli Lilly | 6 | 16 |
| Amazon | 2 | 8 |
| Alphabet | 1 | 4 |
| Pfizer | 11 | 14 |
| Boeing | 5 | 8 |
Climate language thinned out as well. Mentions of "climate" fell 28%, from 78 to 56, with the biggest cuts at JPMorgan (16 to 5), Intel (8 to 2) and Walmart (8 to 4).
If you screen filings for "ESG" or "climate," you'll now undercount how much of this risk companies still disclose. Walmart, Nvidia, Exxon and Pfizer kept the risk and changed a few words.
2. AI infrastructure got its own risk factors
Microsoft, Nvidia and Alphabet each added risks about building and paying for AI capacity. None of these new risks were in their prior-year filings.
Microsoft broke one cloud and AI risk into several, covering infrastructure buildout, outages and capacity constraints, and supply. It now warns that if it overestimates demand, infrastructure could sit underused and lead to impairments. It also says tighter credit or higher rates could limit its ability to execute its infrastructure strategy. And it says it is securing energy resources for AI; mentions of power or electricity went from 4 to 11.
Nvidia added a new heading: "commercial arrangements expose us to counterparty risks." It discloses long-term capacity purchase obligations and financial guarantees, and says it has been asked to help finance customers' and partners' datacenter buildouts. It adds that it hasn't entered into any financing arrangements. Its supply risk heading also changed from "component availability" to "capacity availability."
Alphabet added a heading about disruptions to its ability to access future financing or manage its indebtedness. New text also covers large long-duration commercial agreements, excess capacity it might not be able to redeploy, and counterparties that might not pay.
Investors have spent the past year arguing about whether the AI buildout is overbuilt and how much of it runs on credit, including deals where suppliers help fund their own customers. Three of the biggest companies in that buildout have now put those same concerns in their 10-Ks.
3. Tariffs went from possible to actual
Tariff mentions rose 66%, from 56 to 93. Nine companies increased them and none reduced them.
| Company | Tariff mentions, prior year | Latest year |
| Tesla | 1 | 10 |
| Eli Lilly | 6 | 16 |
| Amazon | 2 | 8 |
| Alphabet | 1 | 4 |
| Pfizer | 11 | 14 |
| Boeing | 5 | 8 |
The tense changed too. Tesla's latest filing says 2025 U.S. trade policy changes, including higher import tariffs and retaliatory measures, "have impacted our supply chain costs." The year before, its risk section mentioned tariffs once.
Export-control mentions rose 16%, mostly because of Nvidia (33 to 38). Tesla mentioned export controls for the first time.
4. AI mentions rose everywhere except Nvidia
Mentions of "AI" or "artificial intelligence" rose 44%, from 308 to 443, and 12 of the 14 companies increased them. Some of the biggest jumps came from outside tech: Pfizer went from 6 to 24, JPMorgan from 8 to 23, and Exxon from 0 to 4.
UnitedHealth gave AI its own heading: "our increasing use of AI presents legal, regulatory and business risks to our operations, reputation and financial results." Last year AI only came up inside its broader information systems risk.
Nvidia went the other way, from 64 mentions to 56. Most of its new language is about capacity, energy, counterparties and export controls.
5. Look at what changed, not the whole section
Risk sections are still getting longer, but only a little. Ten of the 14 grew, and the median change was +3.4%. Microsoft added the most (+17%), and JPMorgan cut the most (-18%) in a near-total rewrite.
Length doesn't say much about direction. Most of each section carries over word for word from the year before, so a sentiment average across the whole thing hardly moves. We got more out of scoring only the sentences a company added and the ones it deleted.
| Company | Length change | Share of text that is new | New sentences, avg score | Deleted sentences, avg score |
| Boeing | -0.6% | 27% | -0.025 | -0.016 |
| Pfizer | +10.4% | 24% | -0.017 | -0.007 |
| Nvidia | +3.5% | 16% | -0.019 | -0.011 |
| ExxonMobil | +9.9% | 31% | -0.013 | -0.002 |
| Tesla | +4.5% | 16% | 0.000 | -0.007 |
| Meta | +3.3% | 11% | -0.008 | -0.004 |
| Amazon | +4.3% | 7% | -0.004 | 0.000 |
At Boeing, Pfizer, Nvidia and ExxonMobil, the added sentences were roughly 1.5x to 6.5x more negative than the deleted ones. Exxon's figures come from a small sample (21 new sentences, 11 removed), so we wouldn't lean on its ratio. Tesla went the other way: its new sentences scored neutral on average, which is more positive than what it cut.
Boeing is a good example of why length misleads. Its section got slightly shorter and went from 28 risk headings to 24, but its new language was the most negative in the group. It removed the risk about the Spirit AeroSystems deal, which has since closed, and added one about keeping a healthy production system and making sure every airplane conforms to exacting specifications.
Amazon is at the other end. It rewrote only 7% of its risk section and deleted almost nothing, which suggests it treats Item 1A as standing legal text rather than something to revisit each year.
6. Two ways year-over-year comparisons go wrong
The first is averaging across subsections. A score averaged over subsections can move just because a company reorganized its headings. Eli Lilly's section was split into 6 subsections one year and 22 the next. Its section-averaged score "improved" from -0.61 to -0.18, while its summed score actually got worse, from -3.66 to -3.95. Microsoft shows the reverse. Going from 39 subsections to 32 pushed its section-averaged score from -0.12 to -0.17, while its sentence-level average barely changed. Scoring at the sentence level avoids both problems.
The second is format changes. Intel's FY2024 10-K didn't use a standard layout; its risk factors sat inside a broader "risk factors and other key information" section. The FY2025 filing uses a standard Item 1A, so a naive diff flags more than half of Intel's risk text as new. JPMorgan has a similar issue, with about 70% of its sentences changed past our match threshold. In both cases most of that is rewording and reorganization rather than new risks. A high rewrite rate is worth a closer look, but on its own it isn't evidence that risk went up.
Run it on your own coverage
We did this for 14 companies. The same structured filings data covers every 10-K in the 2025 and 2026 indices, with sentence-level scores and prior-year links already attached, so the diff runs the same way on a sector, an index or a watchlist. If you want it for your coverage universe, get in touch and we'll run it.
How we did it
The sample is the two most recent annual 10-Ks for each of the 14 companies, excluding amendments (10-K/A). For 11 of them, that's FY2024 (filed January to March 2025) against FY2025 (filed January to March 2026). The other three have non-calendar fiscal years. Microsoft's pair covers the years ending June 2025 and June 2026, and Walmart's and Nvidia's cover the years ending January 2025 and January 2026.
Item 1A text and sentence-level sentiment scores come from Context Analytics' structured SEC filings data. Each filing record links to the company's prior-year 10-K, so we didn't have to match pairs by hand.
Length is the word count of the deduplicated sentences and risk headings in Item 1A.
For sentence matching, we dropped sentences under five words and normalized numbers, so a changed year or dollar figure doesn't register as a change. A sentence counts as edited if it's at least 75% similar to some prior-year sentence, and as new if nothing reaches that threshold. Deleted sentences are prior-year sentences with no match in the latest filing.
Keyword counts are whole-word matches in the deduplicated Item 1A text, where "AI" also covers "artificial intelligence." We cross-checked them against the raw section text, and the direction of change matches for every term reported.
Fourteen mega-caps are a set of case studies, not a market-wide sample. Quotes are taken from the filings as we processed them, so check the exact wording against the original SEC documents before citing them.