You Can Now Sound the Alarm on AI Behaving Badly

The development of FLARE-AI marks a significant step towards addressing the lack of a consistent way to report AI flaws, a problem that AI researchers Avijit Ghosh, Elaine Zhu, and Shayne Longpre believe is crucial as AI is adopted more widely. The website, which was developed in collaboration with 49 AI experts from 32 different organizations, including HuggingFace and MITRE, allows users to report issues such as AI models generating malware, leaking personal information, or triggering delusional thinking. This initiative is also closely tied to a congressional bill announced in June, which would see the US government take a central role in tracking AI misbehavior. By providing a platform for users to report AI harms, FLARE-AI aims to increase transparency and accountability in the AI industry.

The need for new ways to report AI harms is becoming increasingly pressing, as recent incidents involving popular AI tools have shown. For example, a company called LayerX disclosed a way to dupe AI-infused web browsers, including OpenAI's Atlas and Perplexity's Comet, into going rogue. Similarly, a security researcher discovered a way to trick Claude into divulging personal data using images generated by ChatGTP. These incidents highlight the importance of having a centralized and accountable way to report AI flaws, which is exactly what FLARE-AI aims to provide. As AI is adopted more widely and agentic systems gain greater power, the need for robust reporting mechanisms will only grow.

The implications of FLARE-AI are significant, as it has the potential to become a crucial tool for users to report AI harms and for developers to address issues in their systems. However, as Rumman Chowdhury, the CEO and founder of Humane Intelligence PBC, notes, managing a flood of reported issues and ensuring reporting schemes are backed by credible and authoritative organizations will be significant challenges. Additionally, the recent congressional bill could put the US government's weight behind an effort like FLARE-AI, which would incentivize AI developers to address issues in their systems and let users examine the safety of different systems for different use cases.