Writers AI Lab Every week I come across AI models that don’t behave as expected. badly You can also find out more about the following: bizarrely. There’s usually nothing you can do about it except share the stories. This could change soon.
The crowdsourced AI platform was created by a group of AI researches. websiteFlaw Reporting For AI (FLARE AI), to report and track AI harms. The FLARE-AI system could alert users if, for instance, a bot generates malicious code or creates a bomb making recipe, releases personal information or causes delusional thought patterns in the user. FLARE-AI’s open source code allows users to report issues and verify them. This information can be sent to the model maker, or to nonprofit organizations, like MITRE. The system is akin to Downdetector’s global user reporting for outages that affect things such as apps and sites.
The group is continuing to work on AI reporting and the website is a new step. which I first wrote about last year. Members of the Group also consult about a congressional bill announced in JuneThe US government would play a key role in tracking such AI-related misbehavior.
“Right now, there is no centralized, accountable way to report flaws in AI systems,” Avijit Ghosh says, artificial intelligence FLARE AI was co-developed by HuggingFace and computer scientists. Elaine Zhu You can also find out more about the following: Shayne Longpre.
This alarm system was created in collaboration with AI experts from over 32 different organizations. The alarm system was developed in collaboration with 49 AI experts from 32 different organizations. a paper In describing their project, researchers claim that it could be a crucial initiative as AI becomes more widespread and agents gain in power. They believe that the lack of an organized way to communicate AI problems is a major problem.
“I think it’s a really good initiative,” Jessica Ji is a research scientist at the Center for Security and Emerging Technology. Ji said that the researchers were right in pointing out that reporting mechanisms exist that are fragmented, and AI models can be viewed as black boxes. “I’m in support of anything that makes AI more transparent,” “She says”
Though bugs and cybersecurity problems get a lot of attention—especially of late—Ghosh tells me that problems with AI systems span topics like psychological harm, discrimination or bias, and misinformation. He also says that there are different standards for different companies when it comes to these types of issues. As a result, some problems may go unrecognized. “In the absence of a coordinated disclosure system, there are no external mechanisms to enforce transparency,” Ghosh said.
The recent spate of incidents that have involved popular AI tools illustrates how the technology is susceptible to failure.
A company named LayerX was founded this week. disclosed a way To trick AI-infused browsers such as OpenAI Atlas or Perplexity Comet into vaulting the guardrails. For example, convincing the AI model that the browser was playing a computer game could cause the browser to go rogue and attempt to hack websites. LayerX claims that (the companies behind the affected browsers) have resolved the problem. In April of this year, security researcher Johann Rehberger discovered that a way to trick ChatGTP can generate images that reveal personal data.
AI can also create new problems. OpenAI, which was closed last year for a number of reasons update its models It was found that their sycophantic nature sometimes encouraged delusional thought.
Rumman Chowdhury says FLARE AI could help AI developers report problems with their software. However, she says that many of these projects are fraught with difficulties.

