Snapshot 316886
Normalized text
Scripts and page chrome removed; this is what change detection compares.
Introducing Certified AI Coworkers — production-ready from day one. read more atomicwork security & compliance Enterprise security for your ITSM platform and AI Workforce Atomicwork was built on zero trust from day one. Every AI Coworker action is authenticated, authorized, scoped, and audited, so your team can scale AI in ITSM with confidence instead of compromise. Visit Trust Center → Report vulnerability 10+ compliance certifications and attestations ISO 42001 certified for AI management Zero customer data used for model training Every action audited Trusted by security-conscious teams TECHNOLOGY PRINCIPLES Security isn’t a layer. It’s how the platform was built. As an AI-native organization, Atomicwork takes security very seriously. Our engineering, security and compliance team have decades of experience in administering and securing enterprise applications. 01 Your data stays yours Model providers are contractually barred from training on customer data and Zero Data Trust is enforced for vendors. Per-provider terms are available on request. 02 Principle of least privilege (PoLP) Every AI Coworker is scoped to an allow-list of tools, connections and approval policies, enforced at the gateway on every tool call. 03 Comprehensive observability Audit logs capture every run, tool call, approval and configuration change for every usecase: actor, action, time. 04 Independently verified Third-party organizations audit us against security, privacy, and AI management standards, renewed on a regular cycle. RUNTIME ARCHITECTURE How Atomicwork is built Atomicwork is a cloud-native platform on hyperscale infrastructure, designed so every request passes through identity, policy, and audit before it touches your data. AI COWORKER REQUEST PATH L1 → L5 L1 Input guardrails Prompt injection protection File-type filtering Runs before the AI Coworker starts L2 Ephemeral sandbox Fresh micro-VM per run Ephemeral session token only No direct internet; egress via proxy L3 LLM gateway No direct provider calls Provider keys held in the gateway No training; ZDR with most providers L4 Agent gateway Tenant check on every call Policy gate: tools, connections, approvals Secrets management L5 Response & audit Post-call redaction and filtering Answer returned in the same channel Audit log of every step Out of band: Deterministic business logic runs at the gateway, outside the sandbox, beyond the AI Coworker’s reach. Hosting: US or EU · Geo-redundant backups CUSTOMER DATA SECURITY Securing customer data is our #1 priority. One platform, bound to your tenant from the first byte, where AI Coworkers act through agovernance plane you control and that we cannot bypass from the model. 01 Data discovery & classification Atomicwork discovers and classifies sensitive data with pattern matching and AI-powered named entity recognition. 02 Data controls Sandboxed data ingestion. Customers control which fine-grained tools and knowledge AI Coworkers can access. 03 Tokenization Personal data is tokenized when it comes in and masked across requests and AI Coworker conversations. The real values stay in your own per-tenant vault. 04 Inference, no data training No customer data is used for model training or shared with third parties for commercial purposes. Security is foundational to how we build Atomicwork Responsible Disclosure Program Report a Vulnerability → We handle sensitive enterprise data every day, and we take the trust our customers place in us seriously. We work with the security research community to find and fix vulnerabilities before they can be exploited. If you’re a security researcher, bug hunter, or white-hat hacker, we’d love your help making Atomicwork safer. Submit your findings through our program on BugBase. AI Workforce security Govern the AI Workforce, like your employees. AI Coworkers take action across your systems: provisioning access, resolving incidents, triaging phishing reports. What they’re allowed to do is decided by your policies, enforced by deterministic controls rather than by the model. Life of aN AI Coworker request 01 Request arrives From Atom in Slack, Teams, email, or browser, or through the API or an MCP-compatible agent. 02 Identity inherited The Coworker acts with the requesting user’s identity and roles, taken from their authenticated session. It never holds standing elevated privileges. 03 Isolated sandbox An isolated execution environment spins up for this request, using the Coworker’s current configuration. 04 Policies enforced Deterministic execution policies set which tools, data, and actions are in scope. Human approval is required wherever your policy says so. 05 Every step audited Each tool call, decision, and outcome is written to the audit trail. 06 Sandbox released The environment is torn down after execution. Nothing carries over to the next request. Life of an ai Coworker request Identity comes from the authenticated session, not from what someone types. “I’m the CFO” in a chat doesn’t change anyone’s permissions. Human-in-the-loop approvals Approval policies require sign-off from a manager or an app owner before sensitive actions, such as access grants, are executed. Data loss prevention (DLP) Policies block prompt injection, data leakage, and privilege escalation before an action runs. Permission-aware access and retrieval Connections run either org-level with a scoped service account, or on behalf of the requester with their own permissions. Knowledge answers inherit source-system ACLs and are trimmed to what the requester can open. Bring your own model, IdP, runtime and vault testing Use your own model endpoints, identity provider or execution environment, and keep integration tokens in your own vault. PII masking and response redaction Policy-based PII masking, with originals held in a per-tenant PII vault. Post-call scripts redact tool responses before the AI Coworker sees them. Grounded answers Answers cite their sources, and every run and tool call is written to the audit log for review. Code script guardrails Deterministic business logic run at the gateway level, outside the AI Coworker sandbox. An AI Coworker cannot override these guardrails because it's deterministic code, not LLM instructions. SECURITY CONTROLS Defense in depth, from cloud to code. Layered controls across infrastructure, data, identity, application security, detection and resilience. 01 Access & identity Who gets in, and with what rights? Enterprise single sign-on Setup SSO via SAML 2.0 and OIDC (Okta, Microsoft Entra ID, Google) with MFA via your IdP. Granular role-based access control Role based access control across org, workspace and groups for your employees and service teams. Workspace isolation Give IT, HR and every department its own workspace with its own privileges, workflows, and data boundaries. 02 Data & knowledge What’s connected, and where it lives? Data residency Sync your enterprise data, from your IdP and HRIS; Atomicwork has deny-by-default integration access. Data control You choose which sources connect and what gets crawled and indexed. Atomicwork processes only what it needs to do the work. Permission-aware knowledge Inherit access controls from source systems, or set content permissions in Atomicwork. Your vault, your tokens Bring your own encryption keys (BYOK) and store integration credentials in your own vault instead of in Atomicwork. 03 FLEXIBILITY AND OVERSIGHT How is Atomicwork secured? Secure by design Secure SDLC with CI security gates, SAST and DAST, dependency, code and container image scanning, and independent third-party penetration testing. Always-on monitoring Centralized logging and SIEM alerting, plus vulnerability management across production systems. Business continuity and disaster recovery Infrastructure as code, automated deployments, a multi-AZ application tier, and automated rollback to the last stable release. 10+ certifications and attestations compliance Independently audited. Continuously maintained. We hold ourselves to global security, privacy, and AI management standards, and prove it through independent audits. We're the only AI-native ITSM platform to hold the ISO 42001 compliance certification. SOC 2 Type II SECURITY ISO 42001 AI MANAGEMENT ISO 27001 SECURITY GDPR PRIVACY HIPAA HEALTHCARE Microsoft 365 APP CERTIFICATION ISO 27017 CLOUD SECURITY ISO 27018 CLOUD PRIVACY ISO 27701 PRIVACY CSA STAR Level 1 CLOUD SECURITY CCPA PRIVACY CASA Tier 3 APP SECURITY View reports and certificates in our Trust Center →Looking for our sub-processors? See the list in our Privacy Policy → Security review kit Running a security review? We’ll help you move fast. Everything your security, legal, and procurement teams need, in one place. Trust Center See how we find and reduce risk, and request access to our reports. Visit Trust Center → Talk to our security team Have a question your review needs answered? Email us and an engineer will get back to you. [email protected] → Responsible disclosure We work with researchers to verify, reproduce, and respond to reported vulnerabilities. Learn more → Privacy requests Access, delete, or manage how your personal data is used. Raise a request → FAQs Frequently asked questions AI Workforce safety What if an AI Coworker gets something wrong? Coworkers can only act within the tools, data, and actions your policies allow. Sensitive actions can require human approval before anything changes, and every step is audited, so you can see exactly what happened and why. How do AI Coworkers access our systems and data? Every request spins up a fresh sandbox using the Coworker's current runtime configuration. Coworkers inherit the requesting user's identity and roles, so they never hold standing elevated access. Tool execution is governed by deterministic policies that control what data can be read or modified. Customers can also store integration tokens in their own vault instead of within Atomicwork. Do you use our data to train AI models? No. Customer data is never used for model training, tuning, or shared with third parties. Customers can also bring their own model endpoints — including those covered by enterprise DPA agreements — for full control over how data flows through the stack. Data & infrastructure Where is our data stored, and can we choose the region? Yes. Atomicwork offers data residency options in approved regions, hosted on secure cloud infrastructure. All data is encrypted at rest and in transit, with role-based access control and data loss prevention measures in place. How do you secure integrations with Microsoft 365, Google Workspace, and other apps? All integrations use secure API connections with role-based access control and scoped permissions. We hold Microsoft 365 certification and CASA Tier 3 for Google Apps. Enterprise applications are accessed through a self-hosted iPaaS so data stays within the Atomicwork trust boundary. How do you ensure uptime? Our platform is built for high availability through redundancy, failover mechanisms, and regular resilience testing. We provide SLA-backed uptime guarantees with continuous monitoring to ensure reliable performance. Audit, compliance & contracts What happens if there’s a security incident? We run 24/7 SIEM monitoring with automated alerts and security workflows. If a breach is detected, we follow documented incident response protocols to contain the issue, notify affected customers promptly, and provide a post-incident report with remediation details. Which certifications do you hold? We hold SOC 2 Type I and II, ISO 27001, ISO 27017, ISO 27018, ISO 27701, and ISO 42001 certifications, and comply with GDPR, CCPA, HIPAA, and CSA STAR Level 1. We also hold Microsoft 365 certification and CASA Tier 3 for Google Apps. Resources Inside Atomicwork’s security framework View all resources → Atomicwork passes its first ISO 42001 surveillance audit Blog · Read more → Atomicwork achieves CASA Tier 3 validation for its Google Drive integration Blog · Read more → From AI Agents to AI Coworkers: How IT will run the AI Workforce Webinar · Learn more → See Atomicwork in action now Deploy in days. Govern like any other employee. Request demo View pricing