# Google Cloud AI indemnity: who covers copyright claims?

**Google Cloud AI indemnity** answers the question legal teams now ask before any generative deployment: when a service produces content that infringes a third party's copyright, who carries the cost? The company committed to defend its customers against those claims and described the position as a shared destiny with the businesses that adopt its tools. The pledge followed similar moves from Microsoft, Adobe and Canva.

## What the guarantee covers

The protection works on two fronts. The first covers the training data Google uses to build its generative models. If a third party claims that this data infringed its rights, the customer is covered. The second covers content the customer generates through the services, including works created in Duet AI for Google Workspace. Both layers apply to existing agreements, and customers receive them without renegotiating their contracts.

Google's vice-president of legal, Neal Suggs, and its chief information security officer, Phil Venables, framed the announcement as an answer to legitimate customer concerns about copyright exposure when they use AI inside Google products.

## The conditions that apply

Indemnity is not automatic in every scenario. The guarantee assumes responsible use of the tools, and it falls away when a customer intentionally infringes copyright. Vendors also retain the discretion to define acceptable use, which means the practical value of the promise depends on how clearly those limits are written into the terms of service.

The coverage works as a contractual shield rather than insurance against every claim. It addresses intellectual property disputes. Other risks, such as inaccurate output, data protection or regulatory breaches, sit outside the guarantee and remain with the organisation that uses the system.

## Why vendors offer indemnity

The commitments from Google, Microsoft, Adobe and Canva share a commercial logic. Enterprise buyers hesitate when legal exposure is unclear, and a guarantee removes one objection from the sales conversation. Vendors absorb a defined risk to accelerate adoption of products that generate recurring revenue.

The sequence also reflects competitive pressure. Once one major provider promised to cover copyright claims, rivals faced a choice between matching the offer and losing deals on legal grounds. Indemnity moved from a differentiator to an expected term in enterprise agreements.

## What companies should check

A guarantee changes the risk profile only when the organisation understands its boundaries. Legal and technology teams can verify a short list. Does the indemnity cover both training data and generated output, or only one of them? Which products and services fall inside the scope, and which are excluded? What conduct voids the protection, and how much notice is required before a claim? Does the contract require the customer to hand over control of the defence, and who selects the lawyers? Finally, how does the vendor handle a claim that affects several customers at once?

For a business embedding these tools, the indemnity is one input among several. It complements internal controls such as usage policies, access review, output checks and documentation. A contractual promise does not replace governance, and it does not transfer the reputational cost of publishing infringing material.

Who covers the risk also depends on who controls the model, and [the rivalry between Anthropic and OpenAI](/en/ecorner/2023/anthropic-openai-competition.html) shapes the terms every buyer signs.

## Frequently asked questions

**What is AI indemnification?**
It is a contractual commitment in which the vendor defends the customer and covers losses arising from specified intellectual property claims linked to the AI service.

**Does Google Cloud cover both training data and output?**
The commitment addresses both fronts: allegations about the data used to train the models and allegations about content the customer generates through the services.

**When does the guarantee not apply?**
It falls away when the customer intentionally infringes copyright or breaches the acceptable use terms. Other risk categories, such as accuracy or data protection, are not covered.

**What should a business verify before relying on it?**
The scope of products, the conduct that voids the guarantee, the defence process and the interaction with the organisation's own insurance and governance.

## Sources

Google Cloud — Protecting customers with generative AI indemnification — https://cloud.google.com/blog/products/ai-machine-learning/protecting-customers-with-generative-ai-indemnification
Google Cloud — Duet AI data governance — https://cloud.google.com/duet-ai/docs/discover/data-governance
Microsoft — Copilot Copyright Commitment — https://blogs.microsoft.com/on-the-issues/2023/09/07/copilot-copyright-commitment-ai-legal-concerns/
Adobe — Generative AI legal and IP indemnification — https://www.adobe.com/legal/licenses-terms/adobe-gen-ai-user-guidelines.html