Starting October 30, Meta's commercial terms declare that if its AI gets something wrong in your business, that's legally your problem, not Meta's.
Buried in a contract update most businesses will never read line by line, Meta has rewritten the rules for anyone using its AI tools commercially. According to a report from CIO.com, the new terms, which take effect October 30, 2026, state that enterprise customers "are solely responsible for and assume any and all risks" tied to Meta's AI features, even as Meta's own language admits those features "may make mistakes." Meta wrote the admission and the liability transfer into the same clause.
The Business AI Terms of Service go further than a disclaimer. They require customers to "indemnify and hold harmless Meta, its affiliates, employees, and any other agents" against claims, losses, and legal expenses tied to use of its Business Agents, according to a report from WION. That includes attorneys' fees. So if a Meta-powered chatbot tells a customer something that breaks an advertising law or a consumer protection statute, the business that deployed it, not Meta, foots the legal bill for defending against it.
Meta capped its own exposure in the same breath. The terms exclude Meta from responsibility for lost profits, lost revenue, or what the contract calls "consequential, special, indirect, exemplary, punitive, or incidental damages." And the liability doesn't stop at the company signing the contract. CIO.com reported that the new wording also holds IT leaders accountable for anything a partner or contractor does on their behalf while using the tools, stretching the risk down the vendor chain further than most procurement teams expect.
This is worth pausing on, because it's not how every AI vendor handles this. OpenAI's enterprise and API terms include IP indemnification obligations that run the other direction: OpenAI agrees to cover customers against claims that their use of OpenAI's output infringes a third party's intellectual property, subject to a short list of carve-outs like disabling safety features or combining outputs with non-OpenAI products. That's the opposite posture from Meta's. One vendor is telling enterprise customers it will stand behind them on a defined category of risk. The other is telling them they're on their own for essentially everything.
Meta quietly deletes AI training clause from its new business terms Nieman Lab found the original clause in Meta's October 30 business terms, and attorney Matt Topic warned it could hand Meta scraped content worth millions in licensing fees for free. - meta ai training clause scraped business websites - how meta changed ai terms after publisher backlash
Anthropic's approach sits closer to negotiated middle ground: it doesn't train on customer API data by default and offers enterprise agreements with custom terms rather than a single blanket liability shift applied to everyone. None of this makes Meta uniquely reckless. Contract terms get rewritten in AI's favor across the industry as vendors absorb the real cost of models that hallucinate, and Meta's lawyers are simply doing what lawyers do: moving risk off the balance sheet. But the gap between "we'll indemnify you for IP claims" and "you indemnify us for everything" is the gap a business actually has to price in before it signs.
Meta isn't only expanding liability exposure. The same October 30 rewrite also broadens the company's rights over content tied to business accounts and, in an earlier draft, included language permitting its web scrapers and bots to crawl and index business websites more freely. After consultants and reporters flagged the scraping clause as one of three biggest concerns, alongside data ownership and the AI liability transfer, Meta told CIO.com in a statement that it was rethinking that piece, with a Meta public affairs manager saying the crawling language applied only to ads. The liability-shifting language, notably, got no such walkback.
Here's the part that should actually worry founders building on Llama specifically: the model's own license disclaims warranties entirely. Meta provides no guarantee of quality, fitness, or performance for the Llama Materials or anything they output, which means a startup running Llama for free gets zero warranty-based recourse if the model fails in production. Free and open-weight is a real advantage. It is not the same thing as risk-free, and this contract update is Meta spelling that out in writing rather than leaving it implied.
Also read: OpenAI reveals its red-teaming agents found a hardware-level covert channel • Why founders keep failing to vibecode their own LLM inference engine • How to Price AI Agent Seats With Variable Usage That Actually Holds
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.