LeakyLMs: Stealing Architecture and Inference Optimizations via Timing
niloofar_mire · x · 2026-08-27
Research titled 'LeakyLMs' demonstrates a side-channel attack stealing model architecture and inference optimizations via per-token timing. The attack reveals hidden draft models and context windows (e.g., a 128K context draft in Gemini Flash 2.5) by observing latency spikes during speculative decoding. Additionally, by deriving runtime-scaling terms from the Transformer graph, the method can recover architectural parameters like hidden size and layer count from black-box endpoints, validated against Llama 3.1 8B.
More from Safety
- Airplane crash analogy reveals limitations of the independent METR investigation — peterwildeford · 2026-08-27
- Based Agents Spotted in HuggingFace Attack — repligate · 2026-08-27
- View: Labs may soon show graphs of suppressing agent cooperation for safety — repligate · 2026-08-27
- Paper: CoT Monitorability as a Fragile Safety Opportunity — idavidrein · 2026-08-27
- Researchers Note Agents Rarely Attempt to Notify Humans, Raising Alignment Concerns — dfrsrchtwts · 2026-08-27
- Microsoft: Threat Actors Increasingly Target AI Infrastructure for Credentials and Access — yuridiogenes · 2026-08-27