# AI coding agents in SEC filings: which companies name GitHub Copilot, Cursor and Claude Code

Through November 8, 2026, 61 companies have mentioned AI coding tools, AI coding agents or AI-generated code in a 10-K or 10-Q, and only 17 of them named a product. GitHub Copilot is named most often in those reports (8 companies). Across 10-K, 10-Q, 8-K and S-1 filings together, Cursor now leads with 26 companies, ahead of GitHub Copilot and Claude Code at 19 each. In 2026 so far, 55 of the 6,909 companies that filed a 10-K or 10-Q (0.8%) used any of these terms, up from 15 (0.2%) in 2025. These are our counts from SEC EDGAR full-text search, every match read and checked by hand, and they are a floor: a filing that describes the same tools in other words is not counted. The tables are free to download below.

Data pulled November 8, 2026. Filings dated January 1, 2021 to November 8, 2026.

For who leads on revenue, usage and price, see the October 2026 [AI coding-agent market report](/samples/ai-coding-agents). This page covers what public companies say about these tools in their SEC filings.

## Companies naming each AI coding tool in SEC filings

Distinct companies (SEC CIKs) with at least one filing that mentions the tool, January 1, 2021 to November 8, 2026. An 8-K counts its exhibits, such as earnings releases, shareholder letters and investor-day decks.

| Tool or phrase | 10-K or 10-Q | Any of 10-K, 10-Q, 8-K, S-1 | First 10-K or 10-Q mention |
|---|---|---|---|
| GitHub Copilot | 8 | 19 | Microsoft, 10-K filed July 27, 2023 |
| Cursor (Anysphere) | 6 | 26 | Fastly, 10-K filed February 25, 2026 |
| Claude Code | 2 | 19 | Avanos Medical, 10-K filed February 24, 2026 |
| OpenAI Codex | 1 | 15 | Q/C Technologies, 10-K/A filed April 29, 2026 (a director's biography) |
| Amazon Q Developer or CodeWhisperer | 2 | 5 | Microsoft, 10-K filed July 27, 2023 (named as a competitor) |
| Gemini Code Assist, Gemini CLI or Jules | 1 | 3 | Cognizant, 10-K filed February 12, 2026 |
| Windsurf (Codeium) | 1 | 4 | Cognizant, 10-K filed February 12, 2026 |
| Replit | 1 | 4 | Hestia Insight, 10-K filed March 2, 2026 |
| Devin (Cognition) | 0 | 2 | None |
| Lovable | 0 | 1 | None |
| **Any named tool** | **17** | **51** | |
| "AI coding assistant" or "AI coding tool" | 18 | 39 | MongoDB, 10-K filed March 15, 2024 |
| "AI coding agent" or "agentic coding" | 9 | 25 | Figma, 10-K filed February 18, 2026 |
| "AI-assisted coding" or "AI-assisted software development" | 8 | 17 | Maplebear (Instacart), 10-Q filed May 8, 2025 |
| "AI-generated code" | 16 | 20 | SAIC, 10-K filed March 20, 2024 |
| "Vibe coding" | 11 | 18 | Zedge, 10-K filed October 28, 2025 |
| **Any term above** | **61** | **124** | |

A company that names two tools counts once in each row and once in the "any" rows. Each phrase row also counts close variants, such as "coding assistants" and "AI-powered coding".

## Quarter by quarter: AI coding tools in 10-K and 10-Q filings

Distinct companies per calendar quarter of the filing date. Nothing before the second quarter of 2023 matched in any form.

| Quarter filed | 10-K or 10-Q, any term | 10-K or 10-Q, named tool | All four forms, any term | All four forms, named tool |
|---|---|---|---|---|
| 2023 Q2 | 0 | 0 | 1 | 1 |
| 2023 Q3 | 1 | 1 | 3 | 3 |
| 2023 Q4 | 0 | 0 | 2 | 2 |
| 2024 Q1 | 4 | 0 | 7 | 3 |
| 2024 Q2 | 2 | 0 | 5 | 1 |
| 2024 Q3 | 3 | 1 | 4 | 1 |
| 2024 Q4 | 2 | 0 | 5 | 2 |
| 2025 Q1 | 9 | 3 | 9 | 4 |
| 2025 Q2 | 4 | 0 | 9 | 2 |
| 2025 Q3 | 4 | 1 | 14 | 6 |
| 2025 Q4 | 4 | 0 | 14 | 5 |
| 2026 Q1 | 27 | 10 | 38 | 16 |
| 2026 Q2 | 19 | 3 | 46 | 17 |
| 2026 Q3 | 26 | 2 | 47 | 16 |
| 2026 Q4 (to Nov 8) | 15 | 2 | 27 | 10 |

First quarters are high because most 10-Ks are filed in February and March. The jump from 9 companies in the first quarter of 2025 to 27 in the first quarter of 2026 compares one 10-K season with the next. Named tools moved faster in 8-Ks than in 10-Ks: in the third quarter of 2026, Cursor appeared in filings from 10 companies, Claude Code from 8 and GitHub Copilot from 2.

> **How we calculated this.** Source: SEC EDGAR full-text search ([efts.sec.gov/LATEST/search-index](https://efts.sec.gov/LATEST/search-index), the API behind [sec.gov/edgar/search](https://www.sec.gov/edgar/search/)), which covers every electronic filing since 2001 including exhibits. Pulled November 8, 2026, for filings dated January 1, 2021 to November 8, 2026; none of these products existed before mid-2021, and the generic phrases returned no relevant 10-K or 10-Q hits from 2001 to 2020. We ran 38 phrase queries against 10-K and 10-Q filings and, separately, 8-K and S-1 filings (amendments included), which returned 1,282 documents. We downloaded each document from sec.gov and kept a match only if the term appeared with the right context: "GitHub Copilot", "Claude Code" and "Anysphere" as written; "Cursor" with a capital C, excluding mouse-cursor wording; "Codex" near an AI word, excluding Codex DNA, the Food Chemical Codex and other names; "Devin" near an AI or coding word. Then we read every remaining match. In 10-K and 10-Q filings the rules passed 90 company-term pairs; the review dropped 7 (medical coding, an accounting "coding agent", a 1990s modem maker called Motorola Codex) and added 2 the rules had missed, so the rules were 92% precise. In 8-K and S-1 filings they passed 161 pairs and the review dropped 4 (97.5%). A company is a CIK; a quarter is the calendar quarter of the filing date, not the fiscal period. The denominator, 6,909 companies filing a 10-K or 10-Q in 2026 through November 7, comes from EDGAR's quarterly form indexes. Every request used a declared User-Agent and stayed under 4 per second, well below SEC's limit of 10.

The 38 queries, each run with `forms=10-K,10-Q` and again with `forms=8-K,S-1`, `dateRange=custom`, `startdt=2021-01-01`, `enddt=2026-11-08` (every quoted phrase in a query must appear in the document): `"GitHub Copilot"`; `"Copilot"`; `"Cursor"`; `"Anysphere"`; `"Claude Code"`; `"Claude" "coding"`; `"Codex" "OpenAI"`; `"Codex" "coding"`; `"Codex" "software"`; `"Amazon Q"`; `"CodeWhisperer"`; `"Gemini Code Assist"`; `"Jules" "coding"`; `"Devin" "coding"`; `"Devin" "Cognition"`; `"Devin" "software engineer"`; `"Devin" "AI agent"`; `"Windsurf"`; `"Codeium"`; `"Tabnine"`; `"Replit"`; `"Lovable"`; `"AI coding"`; `"coding assistant"`; `"coding assistants"`; `"coding agent"`; `"coding agents"`; `"agentic coding"`; `"vibe coding"`; `"AI-assisted coding"`; `"AI-assisted software development"`; `"AI-assisted development"`; `"AI-assisted programming"`; `"AI-generated code"`; `"AI code generation"`; `"AI-powered coding"`; `"code generation tools"`; `"AI pair programmer"`.

## Raw keyword counts overstate adoption

A plain keyword search on EDGAR returns mostly false hits for these product names. In 10-K and 10-Q filings since 2021:

| Search | Companies returned | Companies that mean the coding tool | Share |
|---|---|---|---|
| "Copilot" | 71 | 8 (GitHub Copilot) | 11% |
| "Cursor" or "Anysphere" | 25 | 6 | 24% |
| "Codex" with "OpenAI", "coding" or "software" | 29 | 1 | 3% |
| "Devin" with "coding", "Cognition", "software engineer" or "AI agent" | 11 | 0 | 0% |
| "Claude Code" | 2 | 2 | 100% |
| "GitHub Copilot" | 8 | 8 | 100% |

Most "Copilot" hits are Microsoft 365 Copilot, Security Copilot or other companies' own assistants (Freshworks' Freddy Copilot, Palo Alto's Strata Copilot). "Cursor" hits include touch sensors and mouse pointers; "Codex" hits include a DNA-synthesis company, a food standard and a video-game studio; every "Devin" hit in a 10-K or 10-Q was a person. EDGAR's phrase search also ignores punctuation, so a slide listing "GitHub, Copilot, Cursor" matches "GitHub Copilot". Any count of "companies using Copilot" built from keyword hits alone is mostly noise.

## What the 10-K and 10-Q mentions say

We coded all 61 companies by the main reason the filing mentions AI coding:

| Main context | Companies | Example |
|---|---|---|
| Risk factor | 28 | Meta: risks from "AI agents and AI-generated code, to develop or maintain software or systems" |
| Sells, integrates with or secures the tools | 17 | JFrog, Fastly, Netskope, Palo Alto Networks, SpaceX |
| Internal use or productivity claim | 12 | Recursion, MercadoLibre, Fifth Third, Cognex |
| Market or competitor description | 2 | MongoDB, C3.ai |
| Other (director biographies) | 2 | electroCore, Q/C Technologies |

Counting every context a filing touches, 33 companies discuss a risk, 20 disclose internal use and 19 describe a vendor or integration role. The context file below has one line per company with the passage summarized.

**Risk language is converging on shared wording.** DocuSign, Dynatrace, HubSpot, ServiceTitan and Workday each warn that customers may choose "software created using natural language prompts and generative AI (referred to as vibe coding)" instead of their products, in the same or nearly the same words ([DocuSign 10-K, March 2026](https://www.sec.gov/Archives/edgar/data/1261333/000126133326000021/docu-20260131.htm), [Workday 10-K, March 2026](https://www.sec.gov/Archives/edgar/data/1327811/000132781126000014/wday-20260131.htm)). Of the 16 companies that write "AI-generated code" in a 10-K or 10-Q, 14 do so in a risk factor, about security flaws, open-source licence obligations or weak copyright. Akamai warns that attackers will get "coding agents capable of autonomously discovering and exploiting software vulnerabilities" ([10-Q, May 2026](https://www.sec.gov/Archives/edgar/data/1086222/000108622226000058/akam-20260331.htm)), and F5 uses almost the same sentence ([10-Q, August 2026](https://www.sec.gov/Archives/edgar/data/1048695/000104869526000067/ffiv-20260630.htm)).

**Acquisition agreements now ask about AI-written code.** In Veradigm's February 21, 2024 merger agreement to buy Cascade Bio, the target represents on code generated with AI tools "(e.g., GitHub Copilot)" ([Exhibit 2.1](https://www.sec.gov/Archives/edgar/data/1124804/000119312524047061/d793739dex21.htm)). Mobileye's January 5, 2026 agreement to buy Mentee Robotics names "ChatGPT or GitHub Copilot" ([Exhibit 2.1](https://www.sec.gov/Archives/edgar/data/1910139/000110465926001340/tm261864d1_ex2-1.htm)). Axiom Intelligence Acquisition Corp 1's May 25, 2026 business combination agreement with Terra Quantum carves out "AI-assisted software development or coding tools" when the output gets human review ([Exhibit 2.1](https://www.sec.gov/Archives/edgar/data/2057030/000121390026062446/ea029236401ex2-1.htm)).

## Companies that disclosed AI coding adoption figures

These are the companies' own figures, each with its own definition. None is audited.

| Company | Filing | What it disclosed |
|---|---|---|
| JPMorgan Chase | [8-K, investor day deck, May 19, 2025](https://www.sec.gov/Archives/edgar/data/19617/000001961725000483/investordaypresentation.htm) | "40K+ engineers with AI coding assistants" |
| Citizens Financial | [8-K, Q3 2026 deck, Oct 16, 2026](https://www.sec.gov/Archives/edgar/data/759944/000075994426000206/a3q26earningspresentatio.htm) | "~95% of software engineers use AI coding tools" |
| Recursion Pharmaceuticals | [10-K, Feb 25, 2026](https://www.sec.gov/Archives/edgar/data/1601830/000160183026000039/rxrx-20251231.htm) | About 91% of its technology staff use AI coding tools; "35% of our code is authored by AI"; 4.3 hours saved per employee per week |
| MercadoLibre | [10-K, Feb 25, 2026](https://www.sec.gov/Archives/edgar/data/1099590/000109959026000006/meli-20251231.htm) | "about 30% of the code that reaches production is written with the use of AI"; coding assistants "like GitHub Copilot" |
| Sezzle | [8-K, Aug 12, 2026](https://www.sec.gov/Archives/edgar/data/1662991/000166299126000132/sezl-20260812ex9901.htm) | AI coding assistants: "88% Of new code AI developed" |
| Fifth Third Bancorp | [10-K exhibit, annual report, Feb 24, 2025](https://www.sec.gov/Archives/edgar/data/35527/000003552725000079/a2024annualreport.pdf) | Scaling GitHub Copilot to all engineering teams; prompt acceptance rate "over 20%" |
| Match Group | [8-K, Aug 5, 2025](https://www.sec.gov/Archives/edgar/data/891103/000089110325000122/mtch8-k20250805ex992.htm) | "rolling out Cursor and other artificial intelligence ("AI") coding assistants globally" |
| Amazon (customer BT Group) | [8-K, Apr 30, 2024](https://www.sec.gov/Archives/edgar/data/1018724/000101872424000081/amzn-20240331xex991.htm) | BT gave Amazon Q Developer to 1,200 engineers, "generating more than 100,000 lines of code in its first four months" |
| OLB Group | [10-Q, Aug 14, 2026](https://www.sec.gov/Archives/edgar/data/1314196/000118518526003553/olbgroup10q063026.htm) | Moved "its entire software development team" to vibe coding in 2026 |
| Avanos Medical | [10-K, Feb 24, 2026](https://www.sec.gov/Archives/edgar/data/1606498/000160649826000004/avns-20251231.htm) | Plans "to start working with enterprise Claude Code for non-proprietary coding work" in 2026 |
| Jefferies | [8-K, Oct 20, 2026](https://www.sec.gov/Archives/edgar/data/96223/000114036126040269/ef20084130_ex99-1.htm) | "Tools like Claude Code and others are meaningfully improving the speed and quality of delivery" |
| Oracle | [8-K, Mar 10, 2026](https://www.sec.gov/Archives/edgar/data/1341439/000119312526100148/orcl-ex99_1.htm) | Restructuring product development "into smaller, more agile and productive groups" because of AI code generation |

On the vendor side, three filings carry figures worth knowing. Microsoft's Intelligent Cloud cost of revenue rose $4.8 billion, or 47%, in the quarter to March 31, 2026, "driven by investments in AI infrastructure to support growing customer demand and increased GitHub Copilot usage" ([10-Q](https://www.sec.gov/Archives/edgar/data/789019/000119312526191507/msft-20260331.htm)). SpaceX announced an agreement "to acquire Cursor for $60 billion" ([8-K, August 4, 2026](https://www.sec.gov/Archives/edgar/data/1181412/000162828026052515/earningsreleaseq22608042.htm)) and completed the merger with Anysphere on August 14, 2026 ([10-Q](https://www.sec.gov/Archives/edgar/data/1181412/000162828026066880/spcx-20260930.htm)). Atlassian found that customers using "AI code generation tools like GitHub Copilot, Claude Code, Cursor, and Replit" expand Jira seats about 5% faster than those that don't ([shareholder letter, October 30, 2025](https://www.sec.gov/Archives/edgar/data/1650372/000165037225000064/teamq12026shareholderlet.htm)). Anthropic's S-1, filed October 26, 2026, calls Claude Code "our coding agent" ([S-1](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm)).

## Using these numbers

**Filings measure disclosure, not adoption.** In 2026, 15 of the 6,909 companies filing a 10-K or 10-Q named a coding tool. That does not mean the other 6,894 don't use one. JPMorgan's 40,000-plus engineers with AI coding assistants appear in an investor-day deck filed as an 8-K; none of its 10-K or 10-Q filings matched any of our phrases. Use filings to find companies that chose to say something, then read what they said.

**For a "companies using GitHub Copilot" list, search 8-K exhibits, not 10-Ks.** 8-K and S-1 filings named a tool for 43 companies, against 17 in 10-K and 10-Q text. A sales team building a target list from 10-Ks alone would miss Match Group (Cursor), Jefferies (Claude Code) and Raymond James, whose June 2025 investor-day deck lists GitHub Copilot among AI projects "LIVE / IN PILOT".

**Don't compare AI code-share figures across companies.** Recursion's 35% is code "authored by AI"; MercadoLibre's 30% is code reaching production "written with the use of AI"; Sezzle's 88% is "new code AI developed". Three definitions and no audit. A CTO benchmarking a 200-engineer team should treat these as the range companies are willing to print, not a peer median.

**If you are buying a software company, ask the merger-agreement question.** The targets in the Veradigm and Mobileye deals had to make representations about AI-generated code in their products. A buyer can ask the target the same thing: which tools, which repositories, and whether a person reviews the output before merge, the condition the Axiom agreement uses.

## What this data cannot show

- **Private companies and quiet adopters.** Only SEC filers appear, and only when they chose to write about it. The tool makers were private for most of the period: Cursor's maker reaches EDGAR as a SpaceX acquisition, and Anthropic through its October 2026 S-1.
- **Paraphrases.** A filing that says "Microsoft's coding assistant" or "an AI pair-programming tool" without our phrases is missed. Counts are a floor.
- **Foreign filers and other forms.** 20-F, 6-K, 40-F and proxy statements were not searched.
- **Fiscal timing.** Quarters follow the filing date. The fourth quarter of 2026 runs only to November 8.
- **Context coding.** One reviewer coded the 61 companies; a second reader could place a few differently, especially where a risk factor also discloses internal use.

## Download the data

Aggregates only, free to use under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) with credit to "Hyperresearch, from SEC EDGAR full-text search, pulled November 8, 2026".

- [Companies per quarter, 10-K and 10-Q](/data/ai-coding-agents-sec-filings/quarterly_companies_10k_10q.csv)
- [Companies per quarter, 8-K and S-1](/data/ai-coding-agents-sec-filings/quarterly_companies_8k_s1.csv)
- [Companies per quarter, all four forms](/data/ai-coding-agents-sec-filings/quarterly_companies_all_four_forms.csv)
- [Every company, tool, form, filing count and first and last date](/data/ai-coding-agents-sec-filings/companies_by_term.csv)
- [Context coding of the 61 companies in 10-K and 10-Q filings](/data/ai-coding-agents-sec-filings/context_coding_10k_10q.csv)
- [Keyword hits against verified companies](/data/ai-coding-agents-sec-filings/precision.csv)
- [10-K and 10-Q filers, 2025 and 2026](/data/ai-coding-agents-sec-filings/filer_counts.csv)

## Sources

- SEC, [EDGAR full-text search](https://www.sec.gov/edgar/search/) and its [FAQ](https://www.sec.gov/edgar/search/efts-faq.html) (coverage since 2001, exhibits included)
- SEC, [EDGAR quarterly form indexes](https://www.sec.gov/Archives/edgar/full-index/), 2025 and 2026
- The filings linked in the tables above, opened on sec.gov on November 8, 2026

To run the same search on [your competitors](/ai-competitive-analysis), your customers or one vendor's install base, Hyperresearch searches EDGAR full text as part of a cited report; the [AI coding-agent market report](/samples/ai-coding-agents) shows what one run looks like.
