Google has revised the wording of its generative AI rules to clarify a significant point for customers: its tools can be used in automated decisions in high-risk domains, provided there is human supervision.
The update matters because automated decisions can shape access to jobs, housing, insurance, social welfare and other areas where the consequences for individuals can be serious. Google says the supervision requirement was already part of its approach, but the new policy language makes the examples more explicit.
What Google Changed
According to Google’s updated Generative AI Prohibited Use Policy, published on Tuesday, customers may use the company’s generative AI tools to make automated decisions that could have a material detrimental impact on individual rights. The condition is that a human must supervise the process in some capacity.
The policy applies to high-risk areas including employment, housing, insurance, social welfare and healthcare. These are domains where an automated output can affect whether a person gets access to an opportunity, service or benefit.
In AI, automated decisions are decisions made by a system using factual and inferred data. A system might, for example, make a decision about whether to award a loan or screen a job candidate.
The previous draft of Google’s terms appeared to suggest a broad prohibition on high-risk automated decision making involving Google generative AI. Google told TechCrunch that customers could already use its generative AI for automated decision making, including in high-risk applications, as long as a human was supervising.
The human supervision requirement was always in our policy, for all high-risk domains, [W]e’re recategorizing some items [in our terms] and calling out some examples more explicitly to be clearer for users.
Why Human Supervision Is The Key Condition
The central boundary in Google’s policy is not whether generative AI can be used in sensitive decision making. It is whether a person remains involved in supervising the system.
That distinction is important because high-risk automated decisions can have consequences beyond convenience or workflow speed. They can influence access to credit, employment, housing, insurance, education-related opportunities, social welfare and healthcare-related decisions.
Human supervision does not remove every concern about AI-assisted decisions, but it is the condition Google has attached to these uses. The company’s position, as reported by TechCrunch, is that the revised terms are intended to make the policy clearer rather than to introduce an entirely new permission.
For customers, the wording clarifies that Google is not treating all high-risk automated decision making as automatically prohibited. Instead, it permits these uses when a human is involved in oversight.
How Google Compares With OpenAI And Anthropic
Google’s approach differs from the rules described for two of its major AI rivals, OpenAI and Anthropic.
OpenAI prohibits the use of its services for automated decisions involving credit, employment, housing, education, social scoring and insurance. That makes its rules more restrictive in several categories where automated decisions can directly affect individuals.
Anthropic allows its AI to be used for automated decision making in law, insurance, healthcare and other high-risk areas, but with additional conditions. The use must be supervised by a qualified professional, and customers must disclose that they are using AI for that purpose.
The comparison shows that leading AI providers are not handling high-risk use in exactly the same way. Google’s updated policy emphasizes human supervision. OpenAI blocks several categories of automated decision making outright. Anthropic allows some high-risk uses, but ties them to professional oversight and disclosure.
Why Regulators Are Watching Automated Decisions
Automated decisions affecting individuals have drawn scrutiny from regulators because of concerns that AI can bias outcomes. The issue is not only whether a system works efficiently, but whether it produces or reinforces unfair results.
Studies show that AI used in decisions such as credit and mortgage application approval can perpetuate historical discrimination. That risk is especially relevant when systems rely on data that may reflect unequal treatment in the past.
Human Rights Watch has called for a ban on social scoring systems in particular. The nonprofit group says these systems threaten to disrupt access to Social Security support, compromise privacy and profile people in prejudicial ways.
Regulators have already started to create obligations for high-risk AI systems. Under the AI Act in the EU, high-risk AI systems, including systems used for individual credit and employment decisions, face the most oversight. Providers must register in a database, perform quality and risk management, use human supervisors and report incidents to relevant authorities, among other requirements.
In the U.S., Colorado recently passed a law requiring AI developers to disclose information about high-risk AI systems and publish statements summarizing their capabilities and limitations. New York City prohibits employers from using automated tools to screen candidates for employment decisions unless the tool has undergone a bias audit within the prior year.
What The Update Means For AI Customers
For organizations using Google generative AI, the updated policy creates a clearer compliance line. High-risk automated decisions are not automatically off limits under Google’s terms, but they require human supervision.
That leaves customers with practical questions about how supervision is designed, documented and applied. The source policy update does not remove the broader pressure around bias, disclosure, oversight or regulatory compliance.
The larger signal is that generative AI providers are defining their own boundaries while governments and local authorities continue to scrutinize automated decision making. Google’s updated terms make one thing plain: in high-risk domains, the company’s permitted-use framework depends on keeping a human in the loop.