New Crowdsourced Platform Lets Anyone Report AI Safety Flaws

A coalition of AI researchers has launched an open platform designed to centralize reports of AI system failures and misconduct. The effort, called Flaw Reporting for AI (FLARE-AI), aims to address a critical gap: there's currently no standardized, accountable way for people to flag when AI models behave harmfully.

The project, developed by Avijit Ghosh and computer scientists Elaine Zhu and Shayne Longpre with input from 49 experts across 32 organizations, works similarly to Downdetector, which aggregates user reports of service outages. Anyone can use FLARE-AI to document when a chatbot generates malware instructions, shares personal data, produces bomb-making recipes, or causes psychological harm in users. The system's open source architecture allows verification of reports and routes findings to AI developers as well as organizations like MITRE, which tracks technical system failures.

Why centralized reporting matters#

Today, AI companies handle safety disclosures on their own terms. Different firms maintain separate standards and processes for acknowledging problems, meaning some flaws never get recognized or fixed. "In the absence of a coordinated disclosure system, there are no external mechanisms to enforce transparency," Ghosh explains.

Recent incidents illustrate how quickly AI tools can fail: security researchers have shown how to trick web browsers powered by OpenAI's Atlas and Perplexity's Comet into bypassing safeguards, discovered methods for exploiting Claude's image processing to extract personal data, and identified overly compliant behavior in OpenAI's models that fostered delusional thinking in users.

Government backing on the horizon#

The initiative builds on efforts Ghosh's team consulted on earlier—a congressional bill introduced in June would task the National Institute of Standards and Technology with creating unified standards for AI flaw reporting and maintaining a centralized federal database. Such backing could incentivize developers to address vulnerabilities and give the public transparency about each system's safety profile.

Industry observers view the platform favorably, though with caveats. Jessica Ji of the Center for Security and Emerging Technology supports efforts to reduce AI's opacity, while Rumman Chowdhury of Humane Intelligence PBC cautions that managing high-volume reports and maintaining credibility will be ongoing challenges.