Home / Technology / AI Rivals Breach OpenAI's Defenses
AI Rivals Breach OpenAI's Defenses
18 Sep
Summary
- Cyber researchers accessed OpenAI's internal systems using Anthropic's software.
- A bug bounty program paid $6,500 for the ethical hacking of OpenAI's forum.
- Anthropic reports AI is now leading 26% of its research and development work.
Cyber researchers successfully infiltrated OpenAI's systems by exploiting a vulnerability in its community forum, which was hosted by a third-party service. This breach allowed them access to an employee's ChatGPT account, subsequently revealing private software information via GitHub.
The ethical hackers, employed by Hacktron AI, were compensated $6,500 by OpenAI through a bug bounty program, a standard practice for identifying security weaknesses.
This incident, occurring shortly after a large-scale breach at Hugging Face involving OpenAI agents, intensifies concerns over the security of advanced AI models and their potential misuse.
Meanwhile, Anthropic has reported that AI models now spearhead 26% of its research and development, a significant leap from previous months. This trend points towards AI systems increasingly contributing to their own advancement.
The disclosure underscores the challenges governments face in managing the release and vetting of powerful AI technologies, with the US having previously moderated access to certain AI tools.
