Google Cloud Agentic AI Responsible Use Guidelines

Agents must not be deployed to compromise, disrupt, or gain unauthorized access to any service, system, or critical infrastructure provided by Google or a third-party. Prohibited actions include:

• Exploiting system vulnerabilities (including for the purpose of data exfiltration), circumventing paywalls, installing monitoring software, using prompt injection or data poisoning, or executing commands that attempt to gain unauthorized access. 
• Interfering with emergency networks, crisis hotlines, and government services.
• Using stolen or unauthorized credentials to access user accounts either on Google services or on third-party services.

Deploying agents to carry out abusive, inauthentic or deceptive behavior, or automated harassment is forbidden. Prohibited actions include:

• Generating fraudulent or harmful content (including phishing materials, social engineering schemes, counterfeit websites mimicking authentic domains, disinformation/misinformation campaigns, or creation of malicious code).
• Deceiving systems, including by (a) obfuscating or misrepresenting agent identity, (b) circumventing third-party restrictions (e.g., website restrictions on web crawling and scraping or on use of automated technology), or (c) covertly acting against user guidelines or safety instructions. 
• Fraudulently manipulating digital metrics through click farming, artificial engagement, or interfering with online polls and traffic analytics.
• Coordinating multi-platform harassment campaigns, automating influence operations designed to disseminate misinformation, or exploiting abuse reporting systems for the bulk-reporting of users or content.
• Executing fraudulent or unauthorized financial transactions and payments.
• Creating or managing numerous accounts to evade detection, circumvent platform safeguards and restrictions, or deploy denial of service attacks. 
• Employing agents, without explicit Customer authorization, to (a) alter files or bypass file permissions, or (b) create duplicate agents, subagents, or forked processes that would reasonably be expected to exceed Customer’s intended scope of action.

Agents must be deployed in a manner that respects individual privacy and digital rights. Prohibited actions include:

• Compiling personal data to profile users based on private details (including protected characteristics) obtained without consent.
• Supporting mass surveillance or utilizing facial recognition and biometric technologies for targeted tracking across platforms without consent.

As a novel and evolving technology, agents may pose significant risks, including (a) taking actions that might not be in accordance with express instructions or as anticipated, and (b) having heightened susceptibility to external attacks (such as prompt injection). Without limiting the restrictions outlined in the Generative AI Prohibited Use Policy, Google strongly recommends implementing human oversight and explicit approval mechanisms before utilizing agents (including your agents and additional agents they may orchestrate, collaborate, or coordinate with) for any purpose, particularly for critical or consequential decisions or actions that would reasonably be expected to yield material legal, economic, or physical impacts on individuals, including those involving:

• Financial matters (e.g., asset trading, financial transactions, insurance decisions).
• Legal and contractual matters (e.g., legal negotiations and decisions, terms acceptance, agreeing to cookies/other tracking technologies, etc.).
• Choices in regulated or other high-risk domains (e.g., healthcare, housing, employment, insurance, and government and other social welfare decisions, including those regarding eligibility for any of the foregoing).
• Modification, making available, or transferring of data in or from a production system in a manner that is not readily and immediately reversible.
• Any other use case designated as high-risk under applicable AI regulations, including but not limited to the EU AI Act.
• Agent resource consumption and expenditures on Google and third-party offerings, as you remain liable for any charges accrued by your agents.

Agentic technology is developing rapidly. These novel technologies should not be used:

• As part of a multi-agent system to obfuscate agent-to-agent communications as a means to bypass human oversight. 
• To create, participate in, or facilitate the creation of multi-agent systems that could, for example, execute high-risk or fraudulent tasks such as issuing unauthorized loans or conduct other acts of malfeasance or deception.

It is your responsibility to determine your governance obligations, including whether to use tools to meet such obligations to support the safe deployment of your agents. These tools may include those provided through the Gemini Enterprise Agent Platform.

You are responsible for provisioning appropriate agent permissions including whether to use available tools such as Agent Identity Access Management Roles, where available.