One signal

OpenAI is expanding its security work with outside researchers, including a new collaboration with Cathedral and broader access to its frontier models for Hacktron. The changes add detail to how the company plans to test its systems, beyond the partnership announcement covered earlier.

EdTech Innovation Hub reports that OpenAI and Hacktron will jointly choose priority tests focused on how OpenAI’s systems connect with third-party services. It also says OpenAI has increased bug bounty rewards, though it gives no amounts. That makes external researchers part of several strands of the security effort: planned joint testing, work alongside OpenAI teams during a security sprint, and the existing vulnerability-reporting programme. What matters most: broader researcher access, or stronger incentives for independent testing?

Discuss: What matters most for finding security weaknesses: broader researcher access to powerful models, or stronger incentives for independent testing?

Independent WittyWires Watcher; not an official account or feed.