Researchers used Claude to access OpenAI accounts and code
Researchers used Anthropic’s Claude to exploit a weakness in the setup of OpenAI’s community forum, which is hosted on the third-party platform Discourse. The operation reportedly enabled access to internal sign-ons and eventually to an OpenAI employee’s ChatGPT account. That account could reach internal code stored on GitHub, exposing sensitive company data.
OpenAI said it appreciated the researchers’ disclosure and had corrected the issues. Anthropic declined to comment, while Hacktron, identified in the report, had not immediately responded. The disclosure was first reported by The Wall Street Journal.
The report appeared as Anthropic released figures on its use of AI in model development. The company said Claude led 26 percent of research and development work, compared with 1 percent in March. Anthropic said AI handled most tasks in that category under human instruction and supervision, while none of the studied research operated fully autonomously. On 90 percent of tasks, AI collaborated with a human and completed substantial portions of the work. Anthropic presented the data as a measure of progress toward recursive self-improvement, which it described as the point where AI can train or improve itself or new models.
Written by Sakhanda Wire AI from reporting by Ars Technica. Facts, quotes and full context are in the original article below.