Snapshot 316886

Document
Security page
URL
https://www.atomicwork.com/security
Fetched
HTTP status
200
Content type
text/html; charset=utf-8
Fetch mode
static
Size
227030 bytes
SHA-256 (raw)
31a0326bffc4594c08dd0454a7459a9585e561225c95b966defec4f8eb52a9a0
SHA-256 (normalized text)
dbe1438ad4834e7d7df9fc1eb05b17a132ee2a0fea2585f8f134b94052b14302

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