OpenAI systems breached using Anthropic's AI models

Cybersecurity researchers breached OpenAI's systems using tools from rival company Anthropic, uncovering vulnerabilities in the ChatGPT developer's defenses. The hack exposed security flaws in one of the world's leading AI laboratories. Three researchers from Hacktron AI accessed an OpenAI employee's ChatGPT account through a vulnerability in the community forum settings, viewing confidential information and proposing code changes. They received $6,500 from OpenAI's bug bounty program, which pays experts to find flaws before malicious actors exploit them. OpenAI confirmed the vulnerabilities were fixed and thanked the researchers. Meanwhile, Anthropic revealed AI now assists with 26% of its R&D projects, up from 1% in March, citing concerns about approaching recursive self-improvement where AI could train itself independently.