Data as of Jul 9, 2026, 08:35 PM UTC

Human Oversight

HRMLESS maintains human oversight of all AI systems proportional to their risk level. No AI system makes autonomous decisions about employment outcomes -- every consequential AI output requires human review before action.

Human-in-the-Loop Requirements by Risk Tier

Each AI system is classified into a risk tier that determines the level of human oversight required.

Critical Risk

AI decisions that directly determine employment outcomes (hiring, termination, compensation changes).

Mandatory human review of every AI-generated recommendation before action
Two-person review for decisions affecting employment status
Complete audit trail with reviewer identity, timestamp, and rationale
Documented override authority with the ability to fully reject AI output
Quarterly review of override rates and patterns

High Risk

AI outputs that significantly influence human decisions about candidates or employees (scoring, ranking, screening).

Human review required before any output is used in a decision process
AI outputs presented as recommendations, never as final decisions
Reviewer must confirm understanding of AI limitations before proceeding
Sampling-based audit of accepted recommendations (minimum 10%)
Monthly review of recommendation acceptance rates

Medium Risk

AI-powered features that support workflows but do not directly affect individual outcomes (summarization, scheduling assistance).

Users informed when content is AI-generated or AI-assisted
Easy mechanism to report incorrect or concerning AI outputs
Periodic human review of output quality and accuracy
User ability to override or edit AI-generated content before use

Low Risk

Internal operational AI (log analysis, infrastructure monitoring, code assistance) with no direct impact on external stakeholders.

Standard monitoring and alerting for system health
Periodic accuracy spot-checks
Clear labeling that content is AI-generated

Decision Authority Matrix

Clear delineation of human authority versus AI advisory role for key decisions.

Decision	Human Authority	AI Role
Candidate screening pass/fail	Human reviewer (mandatory)	Recommendation only
Interview scoring adjustment	Hiring manager	Suggested score with rationale
Bias alert response	AI Governance team	Detection and alerting
Model deployment approval	AI Risk Owner + Engineering Lead	Test results and metrics
Model emergency rollback	On-call engineer (any)	Anomaly detection trigger
Fairness threshold adjustment	AI Governance Committee	Metric analysis

Escalation Triggers

Predefined conditions that require immediate human intervention or escalation.

Bias metric exceeds threshold

High

Automatic alert to AI Governance team. 48-hour triage SLA.

Customer requests human review

High

Route to designated human reviewer within 24 hours.

Model accuracy drops below baseline

Medium

Alert to Model Owner. Automatic fallback if below critical threshold.

Override rate exceeds 30%

Medium

Trigger model review. Potential retraining or recalibration.

Adversarial input detected

Critical

Block request, alert Security team, log for investigation.

Regulatory inquiry received

Critical

Escalate to Legal and AI Governance team immediately.

Override Authority

Any human reviewer can override an AI recommendation at any time without system restriction
Override decisions are logged with the reviewer's rationale for audit purposes
Override patterns are analyzed monthly to identify potential model improvement areas
High override rates (>30%) trigger automatic model review and potential recalibration
Customers and candidates are never penalized for requesting human review of an AI-assisted decision

Documentation Requirements

Audit Trail

Every human oversight interaction is documented to support auditing, compliance, and continuous improvement:

AI system identifier and version that produced the output
Input data summary (without PII) that the AI system processed
AI-generated output or recommendation
Human reviewer identity and timestamp of review
Decision made (accept, modify, reject) with rationale
Any override applied and the justification provided

Request a Human Review

Candidates and customers can request human review of any AI-assisted decision at any time. Contact your account representative or email [email protected] to initiate a review. Requests are triaged within 24 hours.