Meta is preparing to shift legal responsibility for mistakes made by its AI features onto the enterprise customers that use them, according to updates to the company's commercial terms reported by CIO.com.
The changes, slated to take effect on October 30, include language stating that "AI features may make mistakes and may not always produce accurate, complete, or suitable results" and that customers "are solely responsible for and assume any and all risks associated with such AI features and any resulting content." For more context on this story, see our ongoing artificial intelligence updates.
In practice, that means when Meta's AI produces an inaccurate, incomplete, or otherwise faulty result in a business context, the legal exposure falls on the customer — not on the company that built the model.
Accountability 'Turned Upside Down'
The clause has drawn sharp criticism from industry analysts.
"Turns accountability upside down. Meta builds the AI features, admits they 'may make mistakes' and then assigns 'any and all risks' to the business," said Frank Dickson, principal analyst at Dickson Research, in comments to CIO.com.
Mike Wilkes, enterprise CISO at security firm Aikido Security, agreed that the move is perplexing, joining a broader chorus of consultants who flagged three biggest concerns with the new terms: the AI liability transfer, changes to data ownership, and sweeping web-scraping rights.
The updated language also extends responsibility beyond a customer's direct use. Under the new terms, IT executives could be accountable for actions taken by partners or contractors who interact with Meta's AI features on their behalf — a provision that significantly widens the circle of risk.
A Web-Scraping Clause — and a Partial Walk-Back
The most controversial provision beyond liability proved to be a grant allowing Meta to use "automated software such as web scrapers, bots, robots, spiders, crawlers, and user agents" to "access, index, cache, analyze, collect, retrieve, or crawl the domains, web pages, or URLs, and any content available through them, that are associated with your products or services," with the right to "periodically pull updates" from those domains.
Read literally, the clause appeared to let Meta scrape content from domains associated with its business customers regardless of robots.txt preferences — effectively, as critics put it, putting a stake through the heart of the long-standing web standard.
Following an inquiry from CIO.com, Meta moved to clarify. Emil Vazques, a Meta public affairs manager, said in an emailed statement that "the crawling terms we consolidated only applied to our ads and business products and we've clarified that in the terms themselves. It does not change how Meta uses web crawlers. As always, content owners can use robots.txt to signal their preferences."
As of Monday evening New York time, the proposed document had not yet been updated to reflect that clarification, though Vazques said a change was coming.
Data Ownership Terms Set Meta Apart
The terms also grant Meta a broad license to content its enterprise users share through its products. Under the proposed phrasing, using any Meta product grants the company "a license to content that is covered by intellectual property rights like photos or videos you share, post, or upload on or in connection with our Meta Company Products." The license explicitly covers content made available by third parties acting on the customer's behalf, such as advertising agencies or page-management service providers.
Analysts note that this appetite for data distinguishes Meta from other AI vendors. Where many enterprise AI providers now compete on guarantees that customer data will not be used for model training, Meta's advertising-centered business model pushes in the opposite direction — toward maximal rights to analyze and learn from everything its systems encounter.
Why It Matters for Enterprises
Liability allocation has become one of the defining battlegrounds of enterprise AI adoption. Vendors generally want to avoid open-ended responsibility for model outputs, while customers want contractual recourse when an AI error causes business harm — a miscalculated forecast, a flawed document, a bad recommendation acted on at scale.
Most large AI vendors have landed somewhere in the middle: offering service-level commitments, indemnification programs for intellectual property claims, or tiered enterprise agreements that accept some share of output risk. Meta's "solely responsible" formulation stands out for how completely it rejects that middle ground, and it arrives as enterprises are already reassessing risk frameworks around agentic AI systems that take actions, not just generate text.
Security researchers have also pointed out that the liability clause creates a perverse incentive: if the vendor bears none of the cost of model errors, the economic pressure to invest in accuracy improvements, guardrails, and evaluation tooling weakens — while the customer, who typically has the least visibility into how the model works, absorbs the consequences.
The terms are not yet final. Meta could still revise the language before the October 30 effective date — as it has already signaled on the crawling clause after press scrutiny.
For enterprise buyers, the practical takeaway is to review the updated terms line by line before the deadline: assess what AI features are actually in use, whether indemnification exists anywhere in the agreement, and how the partner-and-contractor liability extension interacts with existing agency relationships. Contracts negotiated now will define who pays when AI systems inevitably get it wrong.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
