AI Weekly · Data · updated 2026-07-12
What every US public company is telling the SEC about AI — tracked from the full text of their filings. Free public data, read the way an analyst would: new vocabulary, new risk language, new filers, and the anomalies that become stories.
Number of EDGAR filings containing each exact phrase. "Agentic AI" did not exist in a single filing before mid-2024; it appeared in 814 last quarter.
2026 Q3 excluded (12 days old). Counts are filings, not companies; a filing mentioning a phrase multiple times counts once. Source: SEC EDGAR full-text search, queried 2026-07-12.
What an analyst pass over this data surfaces — each one is a publishable story.
Nine companies used "agentic AI" in SEC filings in 2024. This half-year alone: 561. And it isn't buried in boilerplate — 316 of this year's mentions are in 8-Ks, the form reserved for material events. Companies aren't disclosing agentic AI as a risk; they're announcing it as news.
217 filings in the last 90 days carry the phrase "our use of artificial intelligence" — overwhelmingly in risk factors. The list is no longer tech: Macy's, Peloton, Warby Parker, Wayfair, Build-A-Bear, Equifax, student lender SLM, REIT Kimco Realty — and Marine Products Corp, which builds recreational boats. When a boat manufacturer files AI risk language, the disclosure wave is complete.
Among the 376 first-time "agentic AI" filers: CACI International, which completed its acquisition of ARKA Group in March — adding, in its own words, "space-based sensors and agentic AI-based software" for geospatial intelligence. Unisys, Penguin Solutions, Schrödinger, Nebius and Lantern Pharma also joined the first-timer list — established operators, not startups.
The same first-timer list has a visible tail of microcaps, SPACs and crypto-adjacent vehicles adopting the vocabulary: a SPAC merger prospectus, a Worldcoin-treasury holding company, several shell-adjacent issuers. One Form D in the dataset is literally named "Agentic AI I Mar 2026 a Series of CGF2021 LLC." The word is doing promotional work now — which makes first-mention tracking a due-diligence signal in both directions.
Verbatim passages from the filings, each linked to the original document on sec.gov.
"…increased usage of artificial intelligence and machine learning technologies could expose us to operational, safety, cybersecurity, legal and reputational risks and could adversely affect our ability to compete…"
A powerboat manufacturer now formally warns investors about AI risk — the clearest possible marker that this language has become universal, not sectoral.
"…our use of artificial intelligence technologies presents operational, reputational, data security and legal risks that could adversely affect our business and financial performance, and any failure to effectively leverage artificial [intelligence] technologies in our business could neg[atively affect it]…"
The two-sided disclosure in one sentence: using AI is a risk, and so is failing to use it. That double bind is now standard risk-factor grammar.
"The Company has been testing, refining and implementing initiatives, including Artificial Intelligence ("AI"), and we believe there are meaningful opportunities to better serve our customers and support our colleagues." … "business, legal and ethical challenges related to our use of artificial intelligence in our business operations"
Both halves in one filing: AI as a named growth pillar in the strategy section, AI as a named hazard in the risk factors.
"CACI Completes Acquisition of ARKA Group — Company adds space-based sensors and agentic AI-based software to deliver robust geospatial intelligence…"
A major defense contractor describing an acquisition in agentic-AI terms, in a material-event filing — the vocabulary is procurement language now.
"…prior authorization leverage[s] agentic automation, business orchestration, and purpose-built, fully compliant and governed agents…" · "Announced Acquisition of WorkFusion, a pioneer in AI agents for financial crime compliance."
The RPA incumbent re-describing its whole business in agentic terms, with ARR at $1.853B (+11% YoY) — the pivot, stated to shareholders.
"Proof of Human is cryptographic verification that a user is a unique, living person, not a bot or AI agent. It is foundational infrastructure for social networks, banking, agentic commerce…" — filed by a company holding 283M WLD (~8.3% of Worldcoin's circulating supply), citing a "$6.35 trillion" addressable opportunity.
The other end of the spectrum: agent-economy vocabulary carrying a $6.35T claim in a microcap's material-event filing. Exhibit A for the promo-tail finding.
376 companies used the phrase for the first time this year. A sample across the spectrum:
Verify any name yourself: run the same search on EDGAR ↗