All 100 indicators / Category 9.x: AI-specific
Indicator 9.6
Machine Learning Opacity Trust represents a critical vulnerability where humans develop inappropriate trust patterns toward AI systems whose decision-making processes are fundamentally opaque or incom...
Machine Learning Opacity Trust represents a critical vulnerability where humans develop inappropriate trust patterns toward AI systems whose decision-making processes are fundamentally opaque or incomprehensible. This vulnerability operates through trust transfer phenomena, cognitive closure seeking, and competence substitution. Users unconsciously substitute assessments of an AI system's competence in areas they can evaluate (interface design, speed, technical sophistication) for competence in areas they cannot evaluate (accuracy of hidden decision-making processes, data quality, algorithmic bias). This creates security risks when staff accept AI recommendations without verification, making organizations vulnerable to AI-mediated attacks, model poisoning, and decision manipulation.
Trust transfer phenomena and automation bias create vulnerability when humans develop inappropriate confidence in opaque or biased AI systems without adequate verification processes.
A quick assessment asks these questions; each answer is scored from resilient to exposed.
Systematic verification processes, appropriate skepticism toward AI recommendations, documented AI decision auditing
Some verification processes but inconsistent application, mixed trust patterns
Minimal verification, inappropriate trust in AI systems, acceptance of opacity
Implement mandatory labeling system for all AI-generated communications
Deploy technical controls that automatically tag AI messages with clear identifiers. Establish verification requirements for any unlabeled communications claiming human origin. Create escalation pathway for communications that lack proper AI/human identification.
Develop 20-minute training module teaching employees to recognize and trust their 'uncanny' feelings
Include practical exercises using examples of legitimate vs malicious AI communications. Train employees to use verification protocols when experiencing psychological discomfort with digital interactions. Provide clear scripts for escalating 'something feels wrong' concerns to security teams.
Establish policy requiring verification through separate communication channel for any high-stakes requests
Implement technical system that automatically prompts verification for financial, access, or sensitive data requests. Create simple process for employees to quickly verify human identity through alternative means. Deploy phone or in-person verification requirements for unusual requests, regardless of source authenticity.
Create formal process for employees to report 'uncanny' or uncomfortable digital interactions without judgment
Establish security team response procedure for investigating ambiguous AI-human communications. Implement policy protecting employees who escalate based on intuitive concerns rather than technical evidence. Deploy rapid-response system for employees experiencing confusion about communication authenticity.
Deploy system to monitor patterns in employee responses to AI communications
Establish baseline metrics for normal AI interaction behaviors (response times, verification requests, escalations). Create alerts for unusual patterns that might indicate uncanny valley exploitation. Implement automated detection for communication patterns that typically trigger uncanny valley responses.
Deploy multi-factor verification system triggered by employee uncertainty reports
Implement biometric or behavioral authentication for video/audio communications when requested. Create technical controls that flag communications exhibiting uncanny valley characteristics. Establish secure verification codes or phrases for confirming human identity in suspicious interactions.