You can now sound the alarm about AI behaving badly

writing Artificial Intelligence Laboratory Every week means that I occasionally encounter well-behaved AI models Badly and oddly. Usually, there’s nothing you can do about it, except share those stories with you. But this could change soon.

A group of artificial intelligence researchers set up crowdsourcing Website,Flagging Artificial Intelligence (FLARE-AI), for reporting and tracking ,AI harms. For example, if a chatbot creates malware or a bomb-making recipe, leaks personal information, or triggers delusional thinking in users, FLARE-AI can be used to sound the alarm. The open source code behind the system allows others to investigate the problem and forward reports to modellers, as well as organizations like MITRE, a non-profit that tracks problems with technical systems. It’s a bit like Downdetector, which collects real-time user reports of global outages affecting things like apps and websites.

The website is another step in the group’s ongoing work on AI reporting, Which I first wrote about last year. The group members also consulted on a The congressional bill was announced in Junewhich would lead to the US government taking a central role in tracking this type of AI misconduct.

“Currently, there is no centralized and accountable way to report defects in AI systems,” says Avijit Ghosh, one of the researchers. artificial intelligence Policy researcher at HuggingFace who co-led the development of FLARE-AI with computer scientists Eileen Chu and Shane Longbury.

The alert system was developed in collaboration with 49 AI experts from 32 different organizations. in paper Outlining the work, the researchers argue that their initiative could be crucial as AI is more widely adopted and as agent systems gain greater power. They believe that the lack of a consistent way to report AI flaws is a big problem.

“I think it’s a really good initiative,” says Jessica Gee, a researcher at the think tank Center for Security and Emerging Technology. Ji says researchers are right to note that current reporting mechanisms are fragmented, and that AI models are black boxes. “I support anything that makes AI more transparent,” she says.

Although bugs and cybersecurity issues get a lot of attention –Especially recently– Ghosh told me that issues with AI systems include topics such as psychological harm, discrimination or bias, and misinformation. He adds that different companies have different standards around such issues, which means some problems go unrecognized. “In the absence of a coordinated disclosure system, there are no external mechanisms to enforce transparency,” says Ghosh.

A series of recent incidents involving popular AI tools show how easily technology can deteriorate.

This week, a company called LayerX method revealed To trick AI-powered web browsers, including OpenAI’s Atlas and Perplexity’s Comet, into breaching their firewalls. For example, convincing the AI ​​model behind the browser that it is playing a game could cause the browser to go rogue and attempt to hack a website. (LayerX said the companies responsible for the affected browsers fixed the issue.) In April this year, Johan Ripperger, a security researcher, discovered a vulnerability Way to trick Claude discloses personal data using images generated by ChatGTP.

Artificial intelligence presents new and strange types of problems, too. Last year, OpenAI had to Update their models After I discovered that they were excessively flattering, which sometimes seemed to encourage delusional thinking.

FLARE-AI could be a useful way for many AI developers to implement problem reporting methods with their tools, says Roman Chowdhury, CEO and founder of Humane Intelligence PBC. But she adds that such initiatives are often accompanied by serious challenges.