7
—
OpenAI Community Forum Security Breach: What You Need to Know
Recently, researchers exploited a flaw in the configuration of OpenAI’s community forum hosted by third-party Discourse. This breach allowed them to gain access to internal sign-on and ultimately to OpenAI employees’ ChatGPT accounts, which had access to internal code through GitHub.
OpenAI has since resolved the issue and expressed gratitude to the researchers who reported it. Antropic declined to comment on the matter, while Hactron has not yet responded.
The Wall Street Journal first reported this revelation, coinciding with Anthropic’s release of a new dataset showcasing the institute’s increasing use of AI to develop new models.
According to the report, 26% of Anthropic’s research and development work is now “led” by Claude Models, up from just 1% in March. This shift signifies a significant increase in AI-driven tasks under human direction and supervision.
As AI systems continue to advance, Anthropic highlights their essential role in building the next generation of AI. The company aims to help the public grasp the concept of “recursive self-improvement,” where AI can train and enhance itself or create new models.
This concept raises concerns about AI systems becoming harder to monitor and potentially leading to a loss of human control. However, Anthropic reassures that none of the models they studied operate fully autonomously yet. In 90% of tasks, AI collaborates with humans to perform the majority of the work.
Source: Financial Times. Unauthorized reproduction and redistribution prohibited. Content may not be modified.
Source: arstechnica.com
