2026-10-11 18:00 UTC

llm-security

band: coolmomentum: stable score: 0.003
temperature history

Episodes (7)

Independent review will determine whether Hollow-LLM ghost-weight attacks can reliably fool existing zero-knowledge verification schemes for LLM weights and require stronger integrity checks.
expiredknownscott: low
Independent replication will determine whether SparSEEty can recover generated tokens from sparsity-exploiting LLM serving systems through observable side channels and whether practical serving defenses block the attack.
expiredconvergesscott: medium
The authors of “Stealing Reasoning Traces from Proprietary LLM APIs” claim reasoning traces can be extracted from proprietary model APIs, potentially undermining providers' ability to keep those traces private.
expiredknownscott: low
Independent replication will determine whether the published attack can recover meaningful hidden reasoning traces from proprietary LLM APIs that do not expose chain-of-thought.
expiredknownscott: high
Independent testing will determine whether Sentinel Scan provides meaningful and reproducible coverage of common prompt-injection attacks against LLM applications and agents.
resolvedknownscott: low
The Contrastive Decoding Diffing researchers claim access to base and fine-tuned model logits is sufficient to recover verbatim narrow fine-tuning data without weights or activations, creating a privacy risk for APIs that expose token logits.
expiredconvergesscott: medium
AURA maintainer Ecaterina Sevciuc claims its released behavioral threat cases, heuristic scoring, and schema-validation tools provide reusable social-engineering risk representations for LLM safety pipelines, reducing bespoke threat-library construction without establishing model-level detection accuracy.
seedknownscott: low

Trajectory notes