RiskNet: A large-scale dataset of AI risk incidents from news with alignment and multi-dimensional annotations

Researchers have developed RiskNet, a large-scale dataset documenting AI risk incidents from multilingual news sources, organizing hundreds of millions of reports into structured incident records with standardized classifications. The resource bridges the gap between high-level AI governance principles and empirical evidence of real-world AI harms, providing a foundation for data-driven monitoring and computational analysis of AI safety issues.

RiskNet addresses a critical infrastructure gap in AI governance by transforming dispersed news reports about AI failures and harms into a systematically organized, analyzable dataset. The project reflects growing recognition that regulatory frameworks and governance principles lack grounding in comprehensive empirical evidence about how AI systems actually fail in production environments across diverse sectors and regions.

The proliferation of AI deployments in consequential domains—finance, healthcare, criminal justice, content moderation—has outpaced our ability to systematically track and learn from failures. While anecdotal reports of AI incidents surface regularly in media coverage, no standardized mechanism previously existed to aggregate these signals, identify patterns, or enable longitudinal analysis. RiskNet's multilingual approach and large-scale processing pipeline capture incidents that single-jurisdiction or English-only datasets would miss, revealing a more complete picture of global AI risks.

For the AI safety and governance community, this dataset enables researchers to move beyond case studies toward statistical analysis of incident patterns, failure modes, and risk factors. It supports validation of safety claims, benchmarking of detection systems, and evidence-based policy recommendations. The online exploration platform democratizes access to this knowledge beyond academic researchers.

Looking forward, RiskNet's utility depends on maintaining dataset quality as collection scales, managing potential reporting biases inherent in news coverage, and translating empirical findings into actionable governance improvements. The dataset may reveal whether certain AI architectures, deployment contexts, or company practices correlate with documented harms, informing both regulatory priorities and industry safety investments.

• →RiskNet systematizes hundreds of millions of news reports into structured incident records with standardized AI risk classifications.
• →The dataset bridges the gap between governance principles and documented evidence of real-world AI harms across sectors.
• →Multilingual processing reveals global AI incidents that English-only or single-jurisdiction datasets would miss.
• →The resource enables data-driven analysis of AI failure patterns, enabling statistical approaches to safety research previously limited to case studies.
• →An online platform makes the dataset accessible for browsing, supporting downstream research in safety, governance, and risk analysis.