Researchers have exploited a vulnerability in OpenAI’s community forum, hosted by Discourse, to gain unauthorized access to an internal ChatGPT account belonging to an OpenAI employee. This account had access to sensitive internal code via GitHub. OpenAI acknowledged the issue, stating, “We thank the researchers for contacting us and sharing their findings,” and confirmed that the vulnerabilities have since been addressed. Anthropic, another AI research lab, declined to comment on the incident.
The disclosure, first reported by The Wall Street Journal, coincided with Anthropic’s release of new data indicating a significant increase in the use of its Claude AI model for research and development. The company reported that 26 percent of its R&D work is now led by Claude, a sharp rise from just 1 percent in March. This suggests that AI is increasingly taking on tasks traditionally performed by humans, albeit under supervision.
Anthropic emphasized the importance of this data in illustrating how close the world is to achieving recursive self-improvement in AI, where systems can autonomously train and enhance themselves. However, the company clarified that its models do not yet operate fully independently; they still require human collaboration for 90 percent of tasks.
This incident raises ongoing concerns about the security of AI systems and the potential for loss of human oversight as these technologies evolve.
Follow Cyber Warriors Middle East for further global cybersecurity developments.
