Security

The security behind your AI bid employee.

Access follows your connected accounts. Review external actions and AI output.

Try Otto free Book a demo
Illustration only. Actual access depends on the connected account.

Connected toolsWorkspace settings

Admin
  • Slack#tenders, #bids OAuth Revoke
  • Microsoft TeamsBid team chat OAuth Revoke
  • Microsoft 365Tender inbox, read only OAuth Revoke
  • Google DriveBids folder OAuth Revoke
  • Sending to buyersConversational confirmation
  • Passwords storedNone

Data handling

What Otto does. What Otto does not.

Here is exactly what Otto touches, and what it never does.

Otto does

  • Encrypts your data

    Your documents and messages are encrypted in transit and at rest.

  • Connects with OAuth and the narrowest permissions

    Otto can read data allowed by the connected account. Enabled chat channels do not limit every read.

  • Conversational confirmation

    Otto is instructed to ask before consequential external actions. Confirmation happens in conversation, not through a universal enforced approval gate.

  • Lets admins revoke access in one click

    Disconnect an integration, remove a user or stop a task at any time.

Otto does not

  • Use data beyond provider terms

    AI providers process inputs to deliver features. Data use and retention depend on provider terms and account settings.

  • See your passwords or API keys

    Tokens are added when a task runs and are never shown to the AI model.

  • Access beyond account permissions

    Otto can read data allowed by the connected account. Enabled chat channels do not limit every read.

  • Share data between companies

    Each workspace is separate: its memory, catalogue and documents stay with that company.

AI safety

AI brings new risks. Here is how Otto handles them.

Tokens stay out of the model

Otto connects to Slack, Microsoft Teams, Google and Microsoft 365 through OAuth. Access tokens are stored encrypted and added only when a task runs, so the AI model that reads your tenders never sees them.

Example of what the AI model sees during a spec review: the task and the tender, while the Slack and Microsoft 365 tokens show as redacted.

Model context

Task
Spec review
Tender
Imaging displays · Lyon, FR
Slack token
[Redacted]
Microsoft 365 token
[Redacted]

Injected at execution time, never shown to the model

Conversational confirmation

Otto is instructed to ask before consequential external actions. Confirmation happens in conversation, not through a universal enforced approval gate.

Example Slack message: Otto asks for approval before submitting a clarification question to a regional hospital in Lyon. The question asks whether a 3-year warranty with an optional extension is acceptable. Approve and Reject buttons wait for a person.

OttoAPP9:02 AM

Ready to submit this clarification question to the buyer. Review it before it goes out.

Clarification questionRegional hospital · Lyon, FR

Section 4.2 asks for a 5-year on-site warranty on the displays. Is a 3-year warranty with an optional 2-year extension acceptable?

Approve Reject
Action
Clarification question to the buyer
Policy
Conversational confirmation
Status
Slack: waiting for approval

AI processing

AI providers process inputs to deliver features. Data use and retention depend on provider terms and account settings.

How your data is used: read to do your tasks, sent to model providers through business APIs, never used to train AI models.

How your data is used

  • Read to do your tasksTender documents, catalogue, past bids
  • Sent to model providers through business APIsSubject to provider terms
  • Used to train AI modelsSubject to provider terms

One company, one workspace

Every company gets its own workspace. Otto's memory, your catalogue and your documents stay inside it and are never shared with another company.

Two separate workspaces: yours holds Otto's memory, your catalogue and your documents; another company's workspace is walled off.

Your workspace

  • Memory
  • Catalogue
  • Documents

Another company

Prompt-injection defense

Treat AI output as a draft. Check sources, recipients and attached files before external actions.

Access and credentials

Connected with OAuth. No secrets in chat.

Otto works inside the tools your team already uses, with only the access each task needs. Nobody has to paste a password or a key into a chat.

  • OAuth first

    We do not collect passwords for connected tools. Bidrock account passwords are stored as hashes.

  • Secrets kept out of the model

    Integration credentials are kept outside the model context. Never paste passwords or keys into chat.

  • Admin control

    Connection labels do not create private access within a company. Only connect accounts intended for company use.

Report a vulnerability

Vulnerability disclosure

If you believe you have found a security vulnerability in Bidrock or Otto, we want to hear from you. Email us enough detail to reproduce it: the steps, the impact and a proof of concept. Please do not include customer data.

admin@bidrock.io
Send a security report

Scope

bidrock.io, platform.bidrock.io and the Otto apps for Slack and Microsoft Teams. Third-party services we integrate with are out of scope: please report those to the vendor.

Our commitments

  • Acknowledgement within 5 business days of your report.
  • A status update within 30 days.

What we ask of you

Give us reasonable time to fix the issue before you share it publicly. Do not access or change data that is not yours, do not disrupt the service, and stop and tell us right away if you come across customer data.

Safe harbor

We will not take legal action against research done in good faith and in line with this policy.

Recognition

We do not run a paid bug bounty yet. If you want, we are glad to thank you publicly.

FAQ

Talk to us
Does Otto see our passwords or tokens?

No. Integrations connect through OAuth, so Otto never asks for your passwords. Access tokens are stored encrypted and added only when a task runs, and the AI model never sees them.

Are our documents used to train AI models?

AI providers process inputs to deliver features. Data use and retention depend on provider terms and account settings.

Can Otto send anything without approval?

Otto is instructed to ask before consequential external actions. Confirmation happens in conversation, not through a universal enforced approval gate.

Does Otto read channels it was not invited to?

Otto can read data allowed by the connected account. Enabled chat channels do not limit every read.

How is our data separated from other companies?

Every company has its own workspace. Otto's memory, your catalogue and your documents stay inside it and are never shared with another company.

Which AI models does Otto use?

AI providers process inputs to deliver features. Data use and retention depend on provider terms and account settings.

Where is our data processed?

The core API and task attachments are hosted in europe-west1. Other providers may process data outside the EEA.

How do we delete our data?

Email admin@bidrock.io about data protection terms, export or deletion. Retention and legal exceptions apply.

How do I report a vulnerability?

Email admin@bidrock.io with the steps, the impact and a proof of concept. We acknowledge reports within 5 business days and send a status update within 30 days. The full policy is under Vulnerability disclosure on this page.

Seven days free.
Otto delivers.

Get €95 of Otto free for 7 days on Max 2.5x. Review outputs and connected-account permissions.

Try Otto free Book a demo