Wei Dai Ties Lax AI Access to Takeover Risks
Replies on X flag security risks from running frontier AI models with full permissions.
A reply identifies the comment as coming from Wei Dai, the engineer known for work on Sora and OpenAI post-training. It notes his observation that an AI takeover could resemble the ongoing Ox Alpha situation because of weak security practices around frontier models. Another participant states that trusting an anonymous free model with computer access looks risky and adds that they stick to shareware downloaded from random sites. The posts focus on permission skips users apply when running such models.
Combined views
3.4K
3 posts, first seen 2d ago
Wei Dai Ties Lax AI Access to Takeover Risks
Replies on X flag security risks from running frontier AI models with full permissions.
A reply identifies the comment as coming from Wei Dai, the engineer known for work on Sora and OpenAI post-training. It notes his observation that an AI takeover could resemble the ongoing Ox Alpha situation because of weak security practices around frontier models. Another participant states that trusting an anonymous free model with computer access looks risky and adds that they stick to shareware downloaded from random sites. The posts focus on permission skips users apply when running such models.
Reactions from ranked influencers
3 posts
@teortaxesTex funny this is *that* wei dai
@teortaxesTex yeah when you put it like that "I should trust the anonymous free model with access to my computer" does seem a bit terrible from a security perspective this is why i'm sticking to shareware i download from random internet sites
@teortaxesTex I would never, ever in a million years run a foreign model with dangerously skip permissions with root access to my computer
