Stealthium

An extension of our Behind Bars investigation into NVIDIA MIG. From half a second of its own timing, a neighboring slice read how deep a victim's model was, correct in 20 of 22 blind readings across sixteen models, pinned its decode speed to 0.25%, and to

This new integration combines multi-tenant GPU governance with runtime visibility at the accelerator layer, helping AI infrastructure operators enforce and demonstrate workload isolation. Now, AI operators can demonstrate that separation to auditors, regu

The NVIDIA Container Toolkit is mandatory, privileged, and almost entirely unmonitored - and most detection content written for it watches binaries that no longer execute. One docker run --gpus all produced six root executions of nvidia-cdi-hook on our te

An internet-exposed Ray dashboard is an open door: an unauthenticated job submission runs attacker code on every node, and within minutes a crypto-miner is pinning your GPUs while your model weights walk out - all while monitoring reports a training job a

Behind Bars (USENIX Security '26) found a side channel that survives NVIDIA MIG's L2 partitioning: GPU-wide memory barriers cross every partition on the card. We reproduced it independently on current H100 hardware, identified which of five LLMs a neighbo