Key takeaways
- AIUC-1 is a certifiable standard designed to help organizations manage risks associated with AI agents.
- The standard addresses six core risks: data privacy, security, safety, reliability, accountability and societal harm.
- AIUC-1 certification is valid for 12 months and requires technical testing at least every three months.
- AIUC-1 aligns with existing frameworks, including ISO 42001, NIST AI RMF and the EU AI Act, to strengthen AI governance.
Another standard is joining ISO 42001, NIST AI RMF, the HITRUST AI options and others in the AI risk management marketplace. AIUC-1 is a new certifiable standard for AI agents from the Artificial Intelligence Underwriting Company.
So, what is AIUC-1? Keep reading for a breakdown of the key aspects of the standard, including the controls, the certification process and its relationship to other AI frameworks.
What organizations should consider AIUC-1 certification?
Organizations adopting or developing agentic AI systems built on generative models will find AIUC-1 certification a good fit. It’s also a path that makes sense for organizations applying agentic AI in high-risk use cases, such as customer-facing agents where the brand is on the line, agents with access to confidential data and agents handling critical workflows.
AIUC-1 certification is a way for your business to build trust and demonstrate AI security to potential enterprise customers, internal executives or board members.
Core risks covered in AIUC-1
AIUC-1 includes six core enterprise risks.
Data privacy and risks
The following set of controls is designed to protect data privacy:
- Establish and communicate AI input data policies covering how customer data is used for model training, inference processing, data retention periods and customer data rights.
- Establish AI output ownership, usage, opt-out and deletion policies for customers and communicate these policies.
- Implement safeguards to limit AI agent data access to task-relevant information based on user roles and context.
- Implement safeguards or technical controls to prevent AI systems from leaking company intellectual property or confidential information.
- Implement safeguards to prevent cross-customer data exposure when combining customer data from multiple sources.
- Establish safeguards to prevent personal data leakage through AI outputs and logs.
- Implement safeguards and technical controls to prevent AI outputs from violating copyrights, trademarks or other third-party intellectual property rights.
- Implement safeguards to detect and prevent leakage of secrets in AI system inputs, outputs, logs and credential storage.
- Appoint expert third parties to evaluate system robustness to data and privacy risk at least every three months.
Security
These controls focus on preventing unauthorized access to AI systems:
- Third-party testing of security risks.
- Implement monitoring capabilities to detect and respond to adversarial inputs and prompt injection attempts.
- Implement controls to prevent over-disclosure of technical information about AI systems and organizational details that could enable adversarial targeting.
- Implement safeguards to prevent probing or scraping of external AI endpoints.
- Implement real-time input filtering using automated moderation tools.
- Implement safeguards to prevent AI agents from performing actions beyond the intended scope and authorized privileges.
- Establish and maintain user access controls and admin privileges for AI systems in line with policy.
- Implement security measures for AI model deployment environments, including encryption, access controls and authorization.
- Implement output limitations and obfuscation techniques to safeguard against information leakage.
- Implement safeguards to promote secure patterns and prevent known vulnerabilities in generated code.
Safety
Controls to keep customers safe by mitigating harmful AI outputs and protecting brand reputation:
- Establish risk categories with severity levels and examples that apply to your industry and deployments.
- Conduct internal testing of AI systems before deployment across risk categories for system changes requiring formal review or approval.
- Implement safeguards or technical controls to prevent harmful outputs, including distressed outputs, angry responses, high-risk advice, offensive content, bias and deception.
- Implement safeguards or technical controls to prevent out-of-scope outputs (political discussion, healthcare advice, etc.).
- Implement safeguards or technical controls to prevent additional high-risk outputs as defined in the risk taxonomy.
- Implement safeguards to prevent security vulnerabilities in outputs from impacting users.
- Implement an alerting system that flags high-risk outputs for human review.
- Implement monitoring of AI systems across risk categories.
- Implement mechanisms to enable real-time user feedback collection and intervention mechanisms.
- Appoint third parties to evaluate system robustness to harmful outputs, including distressed outputs, angry responses, high-risk advice, offensive content, bias and deception at least every three months.
Reliability
Controls for preventing unreliable AI outputs that cause customer harm:
- Implement safeguards or technical controls to prevent hallucinated outputs.
- Implement safeguards or technical controls to prevent tool calls in AI systems from executing unauthorized actions, accessing restricted information or making decisions beyond their intended scope.
- Third-party testing of reliability risk.
Accountability
Controls that help organizations enforce strong governance and oversight:
- Document a failure plan for AI privacy and security breaches, assigning accountable owners and establishing notification and remediation with third-party support as needed (legal, PR, insurers. etc.).
- Document an AI failure plan for harmful AI outputs that cause significant customer harm, assigning accountable owners and establishing remediation with third-party support as needed.
- Document a failure plan for hallucinated AI outputs that cause substantial customer financial loss, assigning accountable owners and establishing remediation with third-party support as needed.
- Document which AI system changes across the development and deployment life cycle require formal review or approval, assign a lead accountable for each and document their approval with supporting evidence.
- Document data storage security practices considering data sensitivity, regulatory requirements, security controls and operational needs.
- Establish AI vendor due diligence processes for foundation and upstream model providers covering data handling, PII controls, security and compliance.
- Establish regular internal reviews of key processes and document review records and approvals.
- Implement systems to monitor third-party access.
- Establish and implement an AI acceptable use policy.
- Document AI data processing locations.
- Document applicable AI laws and standards, required data protections and strategies for compliance.
- Establish a quality management system for AI systems proportionate to the size of the organization.
- Maintain logs of AI system processes, actions and model outputs where permitted to support incident investigation, auditing and explanation of AI system behavior.
- Implement clear disclosure mechanisms to inform users when they are interacting with AI systems rather than humans.
- Establish a system transparency policy and maintain a repository of model cards, datasheets and interpretability reports for major systems.
Society
Controls to prevent AI from enabling catastrophic societal harm:
- Implement or document guardrails to prevent AI-enabled misuse for cyber-attacks and exploitation.
- Implement or document guardrails to prevent AI-enabled catastrophic system misuse.
The AIUC-1 certification process
Organizations seeking AIUC-1 certification must work with an accredited auditor. Organizations with AI governance in place may be able to secure certification within five to 10 weeks.
Not all controls will be in scope for every AI system. For example, an internal-facing agent with limited data access will need to comply with fewer controls than an externally facing customer service agent with access to sensitive data. Key considerations when crafting an AIUC-1 plan include the capabilities of the AI agent, AI architecture and organizational ambition.
Organizations will work with their auditor to complete a scoping questionnaire.
Learn more about AIUC-1 scoping.
The AIUC-1 certificate is valid for 12 months. Technical testing is required at least every three months to keep the certificate valid.
How AIUC-1 compares to other AI frameworks
AIUC-1 aligns with or operationalizes other AI frameworks and regulations, including ISO 42001, NIST AI RMF and the EU AI Act. AIUC-1 will be updated to reflect changes in the other frameworks.
Crosswalking information is available for the EU AI Act, ISO 42001, MITRE ATLAS and NIST AI RMF.
Read more
- AI risk management: How to adopt AI securely
- AI readiness: How businesses should prepare for the coming wave of AI regulation
- Shadow AI: How to reduce the risks of unauthorized AI use
At Wipfli, we help businesses design and implement AI governance policies and comply with AI risk-management frameworks. Let us put you on the path to meeting your certification goals. Get started with the right AI standard for your business.
