To expand open-weight models'capacity without proportionally increasing computation, recent architectures incorporate gated parametric memory that retrieves learned values and injects them into intermediate representations. One representative design is Engram, which combines deterministic n-gram lookup with context-dep...
Ariel Fogel, O. Hofman, E. Cohen et al.· 0 citations
This work proposes a training-free method that operates directly on the LLM internal representations, leveraging prior evidence that decision-relevant information is encoded within them, and derives policy-violation scores directly from normalized representations of LLM hidden activations.
Oren Rachmil, Roy Betser, Itay Gershon et al.· arXiv.org· 5 citations· ⚡1
AgenTRIM is introduced, a framework for detecting and mitigating tool-driven agency risks without altering an agent's internal reasoning that provides a practical, capability-preserving approach to safer tool use in LLM-based agents.
Roy Betser, Amit Giloni, Shamik Bose et al.· arXiv.org· 14 citations· ⚡2
RIFT-Bench, a representation-driven methodology for dynamic red-teaming that enables unified evaluations across diverse agentic architectures, is introduced, showing that the approach generalizes effectively to heterogeneous agentic architectures.
Y. Levi, Roy Betser, Amit Giloni et al.· arXiv.org· 2 citations
To test whether the taxonomy supports mitigation, TART, Taxonomy-Guided Actionable Representation, is introduced that makes the taxonomy's key aspects explicit to the planner and downstream sub-agents and consistently improves performance.
Vikas Pahuja, J. Brokman, O. Hofman et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
