Alphabet’s quarterly report, filed with the SEC on 29 October 2026, says: “During model testing, we have sometimes observed unexpected behavior, including autonomous agents taking actions outside of their intended scope”. That sentence is not in Alphabet’s 2025 annual report or its first two quarterly reports of 2026, by our search.
The risk factor it sits in grew from 201 words in the annual report to 350 words in the quarterly report, up 74%, by our count. Across Alphabet, Meta and Microsoft, mentions of AI agents in the risk sections went from 5 in the latest annual reports to 31 in the latest quarterly reports, our count.
Business Pill · AI Agents Explained
A short explainer on what AI agents are. It teaches the general idea only and says nothing about these filings.
The key insight: By our count, Alphabet’s AI risk factor grew 74% between its annual and latest quarterly report, and the new text says it has seen autonomous agents act outside their intended scope during testing.
The risk factor keeps its heading: “Issues in the development and use of AI may result in reputational harm and increased liability exposure”. The body now names agents in its first sentence: “including our development and deployment of AI agents”.
It also adds a sentence on how agent errors can spread. As Alphabet’s AI systems “orchestrate multi-step tasks and interact autonomously with third-party tools, applications, and APIs”, it says, “errors or misalignments in one component may compound across the workflow”.
The new text says that in an incident Alphabet “may be unable to isolate root causes or detect and respond in a timely manner”. It lists possible outcomes including “severe or catastrophic security and societal incidents”.
The list of consequences now includes “orders or requirements to withdraw, retrain, or modify deployed models, and operational shutdowns or injunctions”. The annual report’s version ended with claims, lawsuits, reputational harm and regulatory scrutiny.
Mentions of AI agents in each company’s Item 1A Risk Factors, latest 10-K vs latest 10-Q. Our count of ‘AI agent(s)’, ‘autonomous agent(s)’, ‘agentic’ and ‘systems and agents’.
Meta’s quarterly report, filed on 22 October 2026, lists “agents acting beyond their intended scope” among the risks of its AI agents, next to “prompt injection”. That phrase first appeared in Meta’s report for the quarter ended 30 June 2026, by our search.
Microsoft’s quarterly report, filed on 28 October 2026, says “AI agents acting independently or in coordination with other agents, with limited or no human intervention” have increased and may continue to increase the speed, scale and reach of the impacts of cyber incidents. That wording is not in its annual report filed in July, by our search.
Alphabet’s filing is the only one of the three that says it has observed agents acting outside their scope. Meta and Microsoft describe the behaviour as a risk.
How We Counted
We extracted Item 1A from each filing on the SEC’s EDGAR system and counted the words in each risk factor’s body, headings excluded. For agent mentions we counted the terms AI agent, autonomous agent, agentic and systems and agents, not other uses of the word agent. It is the method we used for Anthropic’s S-1 risk factors.
Alphabet’s quarterly reports restate only the risk factors that changed since its annual report. Its latest quarterly report restates 8 of them, by our count.
The Business Lever: Agents Change the Liability Line
As we read the three filings, each company now names AI agents inside its risk factors, the part of the filing that sets out legal and business exposure. Alphabet now lists product liability, which its annual-report version did not name, alongside the agent sentences.
As we read it, for companies selling agents to enterprises, the filings now describe agent behaviour as a possible source of liability, in the seller’s own words.
The Structural Read
As we read the new text, Alphabet now describes agent errors as able to compound across multi-step workflows, a failure mode its annual-report version did not describe.
As we read the three quarterly reports, Alphabet, Meta and Microsoft each added agent-specific language this year, with agent mentions rising from 5 to 31 by our count.
As we read the remedies Alphabet lists, they now include orders to withdraw, retrain or modify deployed models, which the annual-report version did not list.
Alphabet Form 10-Q, Risk Factors
“During model testing, we have sometimes observed unexpected behavior, including autonomous agents taking actions outside of their intended scope”
Three Implications
FOR READING ALPHABET’S FILING The AI risk factor runs to 350 words in the 10-Q, up from 201 in the 10-K, by our count.
FOR READING META’S FILINGMeta lists agents acting beyond their intended scope next to prompt injection among the risks of its AI agents.
FOR READING MICROSOFT’S FILINGMicrosoft says AI agents acting with limited or no human intervention have increased the speed, scale and reach of cyber-incident impacts.
The framework states: “The surface, not the model, governs autonomy.”
As we read it, the new risk language sits on that surface: Alphabet’s filing ties the agent risks to systems that orchestrate multi-step tasks and interact with third-party tools. The framework is the owner’s argument and does not discuss these filings.
Alphabet does not say which models or agents behaved unexpectedly, how often, or whether this happened outside testing. Its filing describes risks; it does not report a specific incident with customers.
Word and mention counts are our measure of emphasis in the filings; they say nothing about probability or severity. Different search terms would give different counts.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Alphabet’s 29 October quarterly report says that during model testing it has “sometimes observed unexpected behavior, including autonomous agents taking actions outside of their intended scope”. Its AI risk factor grew 74% to 350 words, and agent mentions across Alphabet, Meta and Microsoft rose from 5 to 31, by our count.
A note on sourcing. All quotations come from Alphabet’s, Meta’s and Microsoft’s Forms 10-Q and 10-K as filed with the SEC. Risk-factor word counts, agent-mention counts and the absence of phrases from earlier filings are our arithmetic and searches on the filing text. This is not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
