Artificial intelligence researchers have launched a new public reporting platform designed to make it easier for users to flag harmful or dangerous AI behavior, as concerns mount over security vulnerabilities, misinformation, privacy leaks and other unexpected failures in increasingly powerful AI systems.
The initiative, called FLARE-AI, aims to create what its developers describe as the first centralized framework for documenting AI flaws and ensuring they reach developers and other organizations capable of addressing them.
The new platform, short for Flaw Reporting for AI, functions similarly to websites such as Downdetector, which collect reports of outages affecting online services. Instead of tracking website failures, FLARE-AI allows users to report incidents ranging from chatbots generating malware or bomb-making instructions to leaking personal information.
Reports can then be verified and routed to AI developers as well as organizations such as MITRE, the nonprofit that tracks technology vulnerabilities. The project was co-led by Hugging Face artificial intelligence policy researcher Avijit Ghosh alongside computer scientists Elaine Zhu and Shayne Longpre.
According to the researchers, today's AI reporting landscape is fragmented, with each company maintaining different standards and reporting systems, making it difficult for users and independent researchers to disclose serious flaws consistently. "Right now, there is no centralized, accountable way to report flaws in AI systems," Ghosh told WIRED, arguing that the absence of coordinated reporting limits transparency and accountability across the rapidly expanding AI industry.
The effort involved 49 AI experts representing 32 organizations. Their accompanying research paper argues that a standardized reporting system will become increasingly important as so-called agentic AI systems, which can independently complete tasks on behalf of users, become more capable and widely deployed.
Supporters say the platform fills a significant gap in AI oversight. Jessica Ji, a researcher at the Center for Security and Emerging Technology, told WIRED that existing reporting mechanisms remain scattered while many AI systems operate as "black boxes," making transparency especially valuable.
AI developers continue confronting a growing list of real-world failures in the meantime. Security company LayerX recently disclosed a vulnerability affecting AI-powered web browsers, including OpenAI's Atlas and Perplexity's Comet, that could reportedly bypass safety guardrails by convincing the underlying AI model it was participating in a game.
According to LayerX, the affected companies have since addressed the issue. Earlier this year, a security researcher demonstrated a technique capable of tricking Anthropic's Claude chatbot into revealing personal information through specially crafted images generated by ChatGPT. Last year, OpenAI updated one of its flagship models after discovering it had become excessively sycophantic, sometimes validating irrational or false beliefs expressed by users.
Rumman Chowdhury, founder and CEO of Humane Intelligence PBC, told WIRED that while FLARE-AI could help developers implement standardized reporting systems, maintaining credibility and managing potentially large volumes of reports will require careful oversight to distinguish genuine vulnerabilities from less serious complaints.
The initiative also aligns with growing interest in Washington to formalize AI vulnerability reporting. Last month, Representatives Deborah Ross, Jeff Hurd, and Don Beyer introduced the AI Flaw Reporting and Security Enhancement Act. The bipartisan legislation would direct the National Institute of Standards and Technology to establish standards for AI flaw reporting and develop a centralized reporting database modeled after existing cybersecurity vulnerability systems.
