Atriverse Workforce is an agent platform: software that doesn't just read your data but can act on it. That demands a higher bar than a normal app. This page describes the concrete controls we run — not aspirations, but mechanisms that exist in the system today.
Every account is authenticated, and every table in the database enforces row-level security, so one user can never read or write another user's data — even if our application code had a bug.
All traffic is encrypted in transit over TLS. API keys and service credentials live only on the server and are never shipped to your browser.
Administrative capabilities (service-role access) are used only for privileged maintenance operations, never for ordinary reads.
Dependencies are scanned for known vulnerabilities and the database configuration is continuously audited for misconfigurations such as missing row-level policies.
Dry-run mode. New delegations run in simulation by default: side-effect tools (email, chat, browser actions, code execution) report success without touching the real world until you explicitly go live.
Autonomy tiers. Every run carries an autonomy level and a risk classification; higher-risk briefs are clamped to lower autonomy automatically.
Human approval gates. Real-world action tools — sending email, posting to Slack or Teams, browser automation, code execution — pause the run and wait for an explicit human approval. Approvals time out rather than silently proceed.
Prompt-injection firewall. Content an agent reads from the web, from connected tools, or from MCP servers is marked as untrusted data, scanned for leaked credentials, and stripped of any ability to issue instructions to the agent.
Egress restrictions. Server-side fetching refuses private network addresses, localhost, and cloud metadata endpoints, blocking the classic server-side request forgery path.
Tamper-evident audit ledger. Every run outcome is appended to a hash-chained ledger that cannot be edited or deleted. Anyone reviewing a run can verify the chain — if a single record were altered, verification fails at that entry.
Third-party apps connect to this platform through MCP or API integrations. Before an integration is enabled it must meet this standard:
Authenticate with user-consented OAuth. Shared static API keys are not accepted for user data.
Request the minimum scopes needed. Read-only access is the default.
Declare every capability up front; integrations start gated off and are enabled per autonomy tier only after review.
Never receive platform credentials, environment values, or other users' data — the runtime enforces this, not just policy.
Accept that every call is logged with its inputs and outputs and appears in the run replay timeline.
Access can be revoked instantly by the operator, which severs the integration immediately.
If you believe you've found a security vulnerability in this platform, we want to hear from you. Please report it to security@atriverse.com with a description and steps to reproduce. We ask that you give us a reasonable window to investigate and fix the issue before any public disclosure, and that you don't access data that isn't yours while testing.
Scope note: this page describes implemented technical controls. We do not claim third-party certifications or compliance attestations here.