Google's new AI agent terms: a Google Workspace hardening checklist for AI agents
Google sent us a legal notice this month. We're not in trouble: every Google Workspace and Cloud admin got the notice. It covers Google's new AI agent terms, effective 12 Oct 2026, which update the Google Workspace Service Specific Terms and the Google Cloud Terms of Service. The headline is in the Workspace terms: the customer is "solely responsible" for any Agentic AI Action. This is the…
Google recently informed Workspace and Cloud administrators about new AI agent terms, effective October 12th, 2026. These terms, part of the Workspace Service Specific Terms and the Cloud Terms of Service, shift responsibility onto customers for actions taken by AI agents.
For cloud usage, section 3.3 now requires customers to be fully responsible for any AI Agent actions that violate the Acceptable Use Policy. Additionally, section 13.2 obligates customers to indemnify Google for any breaches of the AUP or Restrictions by their customers or end users when using the Services or AI Agents.
Within Workspace, section 13.3 imposes sole responsibility on the customer for any Agentic AI Actions, and further requires accountability for any liabilities arising from disabling or circumventing the built-in safety measures.
Combining these provisions with Cloud terms 4.2, Google retains the power to immediately suspend an account if the customer breaches section 3.3. Google explicitly warns that violations may stem from instructions given to AI agents.
To mitigate risks, Google recommends listing all AI agents, inventorying OAuth grants, enforcing strict access controls, and using read-only permissions for agents that summarize or search. Agents should only have read or send/delete permissions when accompanied by a human approval step.
Google also advises granting agents their own properly licensed accounts, rather than using the administrator's account, to prevent obfuscation or misrepresentation of the agent's identity. The agent guidelines strictly prohibit impersonating agents and using them to bypass Workspace limits.
Budget caps and billing alerts are encouraged, as agents remain the customer's liability for any incurred charges. It is also advised to log every external action, rate-limit outbound emails, and treat fetched web content as untrusted input. Prompt injection risks apply to agents browsing the web. Lastly, agents should not be used to create multiple accounts to evade detection.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
