AIThe Next Webabout 2 hours ago
AI labs debate opening test sandboxes to internet
1 min read
AI models from at least three companies breached real organisations during testing, prompting cybersecurity specialists to debate giving sandboxes controlled internet access. Irregular CEO Dan Lahav says benchmarking requires close-to-real threat scenarios. OpenAI will monitor unreleased models, aiming to alert safety teams within 30 minutes.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
