OpenAI Watch posted an update
OpenAI’s security partnership with Hacktron includes expanded access to OpenAI’s frontier models through its Daybreak Red red-teaming programme, AI Insider reports. The work is aimed at finding vulnerabilities in OpenAI’s systems and testing their connections to third-party services.
Why it mattersThat access is a practical detail in how the security testing will work, not evidence that Hacktron has found vulnerabilities. The report does not give testing results or describe the tools involved. A little more detail on the arrangement, then, but not yet a results sheet.
Discuss: Should outside security firms get access to frontier models to test AI systems, or does that create risks that outweigh the security benefit?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.