FLARE-AI: Flaw Reporting for AI

The AI flaw reporting ecosystem is fragmented and inefficient: The paper audits 12 existing reporting systems and finds that researchers identifying AI flaws often don't know where to report them, while organizations receiving reports rarely share information with other stakeholders. This fragmentation forces good-faith reporters to submit duplicate forms to multiple places.

Five major design problems plague current AI reporting systems: The audit identifies recurring challenges in discoverability (finding where to report), scope (what counts as reportable), information collection (what data is requested), coordination (sharing between stakeholders), and guidance for liability cases—problems that slow down the identification and fixing of AI failures.

FLARE-AI creates a standardized, interoperable reporting platform: The system uses conditional logic and early classification to collect only triage-relevant information, then allows a single submission to be automatically distributed to multiple developers, coordinators, and incident registries in machine-readable format, reducing reporter burden.

Expert consensus supports a coordinated approach: The research incorporates feedback from 49 experts across 32 organizations (developers, security researchers, and ecosystem coordinators), indicating broad stakeholder agreement that breaking down reporting silos is essential for AI safety.

Streamlined reporting can accelerate AI safety improvements: By lowering barriers to reporting flaws and improving information sharing across the AI ecosystem, FLARE-AI aims to help organizations identify and remediate AI system failures faster, ultimately improving AI safety outcomes.