Trusted Deployers: Open Model Compromise by Ryan Baker

·Nuno Sempere··

Cross-posted from my Sub­stack. The crit­i­cal­ity of a re­sponse to highly ca­pa­ble open mod­els is be­com­ing ur­gent, as An­thropic demon­strates the risk from the re­cent GLM-5.3 re­lease. While loss-of-con­trol is an crit­i­cal topic, this is the topic most likely to cause chaos in the next 6-12 months. If prac­tices don’t change, I’d say there’s a 50% chance that many In­ter­net based ser­vices will have to take them­selves offline for ex­tended pe­ri­ods to ad­dress se­cu­rity be­cause t...

Read full article →

Related Articles

Runtime Assurance Meets Corrigibility: What Happens When an Independent Safety Boundary Blocks an AI Agent? by Idorenyin
Idorenyin · Nuno Sempere · 5h ago
What I learned mapping the AI safety ecosystem — and what’s missing for Africa by tjriliwan
tjriliwan · Nuno Sempere · 5h ago
Will AI Agents Pay to Avoid Killing Animals? by Jonah Woodward
Jonah Woodward · Nuno Sempere · 15h ago
US Average Gas Price Is $4.3800 or more on October 5 2026?
Jack · Manifold Markets · 1d ago
Wastewater monitoring as an early warning for pandemics (A Happier World video) by Jeroen Willems🔸
Jeroen Willems🔸 · Nuno Sempere · 1d ago