$$\mathbf{L^2\cdot M = C^2}$$ Large Language Models are Covert Channels

Documenti analoghi

• CatBack: Universal Backdoor Attacks on Tabular Data via Categorical Encoding
  di: Tajalli, Behrad, et al.
  Pubblicazione: (2025)
• EmoBack: Backdoor Attacks Against Speaker Identification Using Emotional Prosody
  di: Schoof, Coen, et al.
  Pubblicazione: (2024)
• The SkipSponge Attack: Sponge Weight Poisoning of Deep Neural Networks
  di: Lintelo, Jona te, et al.
  Pubblicazione: (2024)
• Towards Backdoor Stealthiness in Model Parameter Space
  di: Xu, Xiaoyun, et al.
  Pubblicazione: (2025)
• Let's Focus: Focused Backdoor Attack against Federated Transfer Learning
  di: Arazzi, Marco, et al.
  Pubblicazione: (2024)