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 developer version: what the terms say, then what to actually change.
The three clauses that matter
- Cloud terms 3.3 now restricts using "the Services or any AI Agent ... in violation of the AUP" (the Acceptable Use Policy).
- Cloud terms 13.2: you indemnify Google for "Customer's or an End User's use of the Services or AI Agents in breach of the AUP or Section 3.3 (Restrictions)".
- Workspace terms 13.3: "Customer is solely responsible for: (i) any Agentic AI Action", and 13.3(b) adds "Customer is solely responsible for any liability arising from disabling or bypassing any of these safeguards" (the human confirmation safeguards).
Combine 3.3 with Cloud terms 4.2, which says Google "may immediately Suspend" if "Customer is in breach of Section 3.3 (Restrictions)", and the practical risk is clear: a misbehaving agent can get the account suspended. Google's account help page says it directly: "Violations may result from instructions you give to an AI agent."
The 5 guardrails
- List your agents. Inventory OAuth grants. Personal accounts: Google Account > Security > third-party connections. Workspace: Admin console > Security > Access and data control > API controls, then review and restrict third-party app access.
- Read-only access. Use read-only Gmail and Drive scopes for agents that summarize or search. Grant send or delete only to a flow that has a human approval step.
- A human approves sends. If a tool asks for human confirmation, keep it on. SST 13.3(b) puts bypass liability on you.
- A separate account. Give agents their own properly licensed account or service account, not the founder's admin account. The agent guidelines prohibit "obfuscating or misrepresenting agent identity". Don't use it to share files around Workspace limits.
- A spending cap. Budget caps and billing alerts. The agent guidelines: "you remain liable for any charges accrued by your agents."
Also worth doing
- Log every external action: timestamp, tool, target, approver. An append-only file or table is enough.
- Rate-limit outbound email and posting in code, not in the prompt. "Spam, phishing, or malware" is on the Generative AI Prohibited Use Policy list.
- Treat fetched web content as untrusted input. Prompt injection is a real risk for any agent that browses.
- Don't use agents to multiply accounts. "Creating or managing numerous accounts to evade detection" is on the agent guidelines' prohibited list.
Not legal advice. Read the terms for your own setup.
The flip side: AI is also how people find your product now. Developers and buyers ask ChatGPT or read Google AI Overviews, so AI visibility in AI search is worth tracking too. That's what I work on at Serpair.
Originally published on Serpair: Google's new AI agent terms, explained.
How are you using AI in your business right now? Did Google notify you too? Have you changed anything since, and what guardrails do you have in place? Drop your setup in the comments and I'll fold the best ideas into this checklist.
