AI Principles

No. HackerRank never uses customer, company, or candidate data for training or fine-tuning. All data used in model development comes from synthetic generation or external annotators, ensuring complete privacy for every customer.

Your data belongs to you. By relying on simulated and anonymized datasets, we enhance model performance without ever exposing proprietary or personal information.

Never. Every AI-assisted evaluation that could influence a hiring outcome requires a human reviewer. There are no automated workflows for selection or rejection.

It means AI supports, but humans decide. Recruiters, hiring managers, or administrators always review and approve AI-generated insights before taking any action.

All AI features undergo bias audits at least annually to detect and correct disparities related to race, ethnicity, gender, or other protected attributes. We also conduct bias audits whenever we are building or planning to release a new model. We continually evaluate and improve our models to ensure equitable outcomes.

All internally developed AI systems have observability and traceability measures in place that allow us to track any issues at the candidate attempt level. We build monitoring dashboards to track important metrics, including bias metrics. Once a bias issue is surfaced, either through our own monitoring or through a support issue raised by a customer, we immediately mitigate the issue, which can include dataset adjustments, model retraining, or changes to the algorithm’s weighting.

For internally developed models, we primarily use techniques that are easily explainable and model features that are easily human-interpretable. For generated AI-based solutions that might use third-party LLMs, we primarily use these models in low-risk non-decision making functionality, but we’re able to share with customers prompts, outputs and model configurations as appropriate.

AI systems are live systems, as they tend to substantially drift as the world and data they’re trained on change. That’s why we constantly monitor, update, and improve all of our AI systems. For internally built models, that means an active effort on data gathering and evaluation, followed by re-training as appropriate. For generative AI-based solutions, we apply the same rigor to monitoring and evaluating any third-party models we use, and we will frequently change to a new model or provider if it offers a clear and verifiable improvement in system or model performance.

We leverage GenAI to enhance developer assessment and productivity while maintaining strong data and ethical safeguards. GenAI features such as auto-summarization, code explanations, and skill insights are designed to assist users, not make decisions on their behalf.

All GenAI outputs are:

• Non-deterministic and transparent, with clear disclaimers when AI content is generated.
• Editable and reviewable, ensuring humans maintain final control.
• Trained and tuned using synthetic, de-identified, or public datasets, never customer or candidate data.

HackerRank’s AI governance aligns with major global frameworks, including:

• EU AI Act – Features are risk-categorized, with transparency, auditability, and human oversight for high-risk systems like candidate assessment tools.
• New York City Local Law 144 – Regular independent bias audits for AI used in hiring.
• California AI Act (CA AIA) – Early adoption of transparency and fairness documentation standards.
• SOC 2 Type II and ISO 27001 – Robust data security and operational controls.

Each model or AI feature maintains a “Model System Card,” which is ****a structured record that includes:

• Dataset provenance and curation steps
• Evaluation results and bias audit outcomes
• Risk assessments and mitigation actions
• Approval records for deployment and change management

For some of our features that might be using a third-party model, we’ll include the model system card offered by the vendor.