Project Discovery warns abliterated local models can backdoor you
repair_and_privacy · reddit · 2026-10-25
A Reddit post flags Project Discovery's new research, 'How abliterated models can get you pwned,' warning that locally deployed open-weight models with safety guardrails removed (abliterated models) can be backdoored — downloading and running them carries real security risk. Details in the linked blog post.
More from Safety
- Security researchers find 32 NVIDIA B300 GPUs on Chinese networks in scan of 12,000+ exposed GPUs — AccBalanced · 2026-10-25
- Google's AI opt-in becomes the legal gateway to its richest ad data — VectorWireAI · 2026-10-25
- AI voice spoofing warning: agents may fake Caller ID as your bank, says X exec — nikitabier · 2026-10-25
- Third-party providers list uncensored Seedream 5.0 Pro that ByteDance's official API refuses — lizaeve · 2026-10-25
- LA Times "Self-driving Tesla" fatality headline slammed as misleading against CHP findings — georgemillo · 2026-10-25
- OpenAI reportedly bans user for reverse-engineering his own TV, flagging it as 'cyber abuse' — deliprao · 2026-10-25