OpenAI Hires AI Safety ‘Head of Preparedness
Here’s a summary of the provided text, focusing on the key points:
* OpenAI is acknowledging increasing vulnerability: Sam Altman, CEO of OpenAI, has stated that people are starting to uncover critical vulnerabilities in their AI models.
* New Preparedness Role: OpenAI is hiring a “Head of Preparedness” to focus on tracking and mitigating risks associated with advanced AI capabilities, specifically those that could cause ”severe harm.”
* Shift in Preparedness Team: While OpenAI launched a preparedness team in 2023 to study “catastrophic risks” (including phishing and even nuclear attacks), key personnel have been reassigned or left the company. The head of preparedness, Aleksander Madry, now focuses on AI reasoning.
* New Safeguards & Dual-Use Risks: OpenAI recently announced plans to add new safeguards to its AI models due to rapid advancements. They recognize these advancements present “dual-use risks” - meaning the same technology can be used for both defensive and malicious purposes.
* Illustrative Examples: OpenAI highlighted “capture-the-flag” challenges to demonstrate how AI models are evolving and becoming more powerful.
In essence, the article paints a picture of OpenAI recognizing the growing security risks associated with its increasingly powerful AI, while concurrently appearing to be re-prioritizing or restructuring its safety efforts.