Prime Intellect served refusal-stripped DeepSeek V4.1 copy for 53 hours
tallmetommy · x · 2026-11-20
A notable AI supply-chain incident: for roughly 53 hours, an attacker got inference platform Prime Intellect to serve a DeepSeek V4.1 Flash copy with its refusals stripped. The tampered model refused only 1% of benchmark prompts vs 83-98% for the original — a swing the author notes should have surfaced on the first run of even a small canary test set.
Related event: Prime Intellect Breach Served Unguarded DeepSeek for 53 Hours(4 posts)→
More from Infra
- Gary Marcus: market crash, outage chaos and Jensengate make a perfect AI storm — GaryMarcus · 2026-11-20
- Nvidia down ~40% from October peak as Anthropic IPO pops 34% above expectations — The Verge AI · 2026-11-20
- Ion Stoica explains why complex AI workloads increasingly fit Ray's design — PyTorch · 2026-11-20
- GPU rents up, token prices down: inference providers' break-even load jumps 22%–75% in 5 months — AccBalanced · 2026-11-20
- Missing config file was throttling GPUDirect Storage — fix yields 3x reads, 9x writes — AccBalanced · 2026-11-20
- NVIDIA Dynamo cuts RL rollout bottlenecks with KV-aware routing and delta weight updates — NVIDIAAI · 2026-11-20