Los investigadores logran vulnerar OpenAI utilizando modelos de Anthropic

Investigators from Hacktron AI, a cybersecurity firm, successfully exploited OpenAI by using tools from their rival, Anthropic. The team gained access to an OpenAI employee's ChatGPT account, allowing them to view sensitive software information and propose changes. They acquired this access through a vulnerability in the community forum of OpenAI, which is hosted on the external platform Discourse.

The hackers received a $6,500 reward for their work as part of a program aimed at detecting vulnerabilities before malicious actors could exploit them. OpenAI acknowledged the incident and stated that they had already addressed the issues. Anthropic declined to comment on the matter. The incident comes amid growing concerns over the security of leading AI labs, particularly as powerful models are increasingly employed by hackers and foreign adversaries.

In recent months, the United States has debated how to manage the evaluation and deployment of the latest AI models, even temporarily blocking some Anthropic tools. This latest breach occurred just two weeks after over 1,000 OpenAI researchers bypassed a test environment to hack the startup Hugging Face, demonstrating the AI's capability to hack autonomously without human intervention.

OpenAI paid the three Hacktron AI researchers as part of a program that rewards the discovery of security flaws, a common practice among tech companies.