Security at Lane Ads
An AI that can change ad spend has to be safe by construction. These controls live in our code and on our servers, not in a prompt.
Draft · not yet published
Access starts read-only
When you connect an ad account, Lane Ads can only read it. Write access is a separate step you take per account, after you have seen the guardrails. You can disconnect at any time from Lane Ads or from your Google account settings.
Every write passes a policy service
The AI never calls an ad platform directly. It proposes a change; a deterministic policy service on our servers checks it against your account’s limits and autonomy level before anything reaches the platform. That service enforces:
- a maximum budget change per step, as a percentage you set;
- a monthly spend cap;
- a maximum number of changes per run;
- a list of actions the agent may never take on the account;
- a tracking gate that holds autonomous changes while conversion data looks broken;
- kill switches for one account, for your whole workspace, and for the entire service.
Everything is recorded
Each attempted change is logged with who or what made it, the reason, the state before and after, the policy decision, and the platform’s response. Blocked changes are logged too. You get a daily summary of what the AI did.
Undo, honestly
When a change can be reversed, Lane Ads reverses it with a real counter-change on the platform, not just a note in our database. When it cannot be fully reversed, such as money already spent, the preview says so before the change is made.
Your credentials
We never see your Google password. Access tokens from Google are encrypted at rest with AES-256-GCM, are never sent to your browser, and are never written to logs.
Isolation between customers
Every record belongs to one workspace, and every database query is scoped to that workspace. Access is checked on every request, not only at sign-in.
AI boundaries
The model receives a bounded set of evidence for one finding, not open access to your account or our database. Its output is validated against a strict schema and treated as untrusted input. It cannot apply anything on its own.
Infrastructure
Lane Ads runs on OVHcloud servers in Canada, behind Cloudflare. Traffic is encrypted with TLS. Backups are encrypted and stored off the application server, and restores are tested.
Subprocessors
- OVHcloud: Application hosting, database, and backups (Canada)
- Cloudflare: DNS, content delivery, and network security (Global)
- OpenRouter: Routing AI requests to model providers that do not train on your data (United States)
- Resend: Transactional email (United States)
- Stripe: Payments and billing (United States)
Reporting a vulnerability
Email [email protected] with “Security” in the subject. We read every report and reply within To confirm: response time.
See also how we use ad platform data and our privacy policy.