Recent research demonstrates that current security monitors often fail by confusing injection exposure with an agent's actual commitment to perform an unsafe action. The author proposes a double-stratified evaluation method that separates exposure from behavior, significantly improving the ability to predict unsafe tool calls up to two steps in advance.
Researchers have proposed LOTUS, a framework for hallucination detection that uses multilinear low-rank decomposition to extract signals from hidden LLM layers. By analyzing peak activations and cross-layer patterns, the method identifies incorrect responses more effectively than existing pooling-based techniques.
Researchers have developed TraceQuery, a sparse decoding framework that reduces the computational cost of diffusion large language models (dLLMs). By analyzing denoising trajectories, the method selectively recomputes only necessary token representations rather than performing redundant operations, maintaining high accuracy with fewer computations.
Researchers have introduced the Boa Compressor, an online tool utilizing a Mamba state-space model to compress long-form context into shorter representations in linear time. By distilling knowledge from full-context LLM interactions, this method allows for efficient per-query inference without the massive latency overhead of transformer-based compression.
Researchers introduced Recursive Semantic Divergence (RSD), an unsupervised metric that measures and penalizes inconsistency in LLM agent trajectories. By applying RSD as a reward signal during fine-tuning, models exhibit significantly more stable behavior and higher success rates across tool-use benchmarks.
This research introduces a verification-based framework that refines multimodal emotion reasoning in AI agents. Instead of retraining models, the system identifies specific component failures and applies targeted prompt updates to improve performance.
CISA, together with the Canadian Centre for Cyber Security and other partners, has released joint guidance aimed at helping organizations track and control how much they spend on tokens when using outside AI (described as 'super intelligence') services. The agencies point out that letting token usage run unchecked can drive up costs unpredictably, slow or interrupt service, and leave systems vulnerable to attacks designed to drain a victim's token budget or wallet. To counter this, the guidance lays out steps covering how tokens get allocated, how workflows are built for efficiency, how prompts are written, how caching and reused context can cut waste, how data is trimmed before it's sent to a model, and how usage should be tracked over time.
Red Hat, IBM, and Google announced llm-d 1.0, a Kubernetes-native distributed inference serving system that moves beyond proof-of-concept performance tricks into full production readiness. The release adds multi-tenant quality of service, zero-downtime operations, and agent-aware serving so the same endpoint can handle both human users and autonomous agents efficiently. In one test, llm-d's prefix-aware router captured 97% of available cache reuse on an agentic coding workload, and IBM/Red Hat ran 3,000 concurrent software-engineering agents for 5 to 10 times less cost than using the same model as a hosted service.