En Passante №25

The through-line

Agents are moving beyond chat into specialized interfaces and domain-specific tasks. GitHub pushes past conversation with canvases; Cloudflare automates security configuration; AWS patterns multi-account access; fuzzing and forecasting agents optimize for their particular problem spaces rather than general reasoning.

Issue signature

The Passing Shot

Research & Model Engineering

When The Gateway Swaps Tokens, User Identity Stops Travelling

AWS's four-account AgentCore reference architecture puts Cedar authorization at the gateway hop, which means the one config flag the sample leaves off is the one holding the whole boundary up.

335 words · 1 source · verified

The Wire

PAWS: Policy-driven Agentic World Simulation arXiv cs.AI
Pistis Technical Report arXiv cs.AI
BaseCamp --- An Agentic AI Framework for Automating DNA Sequencing Data Pipelines arXiv cs.AI
DEEPO: Dual-Entropy Enhanced Policy Optimization for Hallucination in MLLMs arXiv cs.AI
TWIST: A Proposed Benchmark for Intervention Quality in Conversational Memory, with a Human-Validated Draft-Alignment arXiv cs.AI
Adversarial Closed-Loop Curriculum for Evolving Role-Playing Agents arXiv cs.AI
Training Object Permanence in World Models arXiv cs.AI
Beyond Surface Style: Aligning Multi-Turn User Simulators with Behavioral Consistency arXiv cs.AI
Driving Epidemic Models with AI Agents: the Epydemix Agent Framework arXiv cs.AI
Progressive Skill Discovery as Access Control for Tool-Using LLM Agents: Structural Governance through Role-Scoped Capability Delivery arXiv cs.AI
Reinforcement Learning with Verifiable Rewards for Small Search Agents arXiv cs.AI
Agent Memory with Episodic Retrieval for Financial Decision-Making arXiv cs.AI
Learned Cross-Task Relationships in Multi-Task Models arXiv cs.AI
RECLAIM: Can Agents Reproduce the Claims of Machine Learning Papers? arXiv cs.AI
Human-AI-Powered Hypothesis Testing: Cost-Aware Selective AI Scoring and Sequential Human Escalation arXiv cs.AI
Forecast-Dojo: Replayable Environments for Benchmarking and Training LLM Forecasting Agents arXiv cs.AI
Control the Harness, Control the Cost: Routing and Governing AI Coding Agents in the Enterprise arXiv cs.AI
PFArena: Benchmarking Language Models for Protein Modification arXiv cs.AI
From Static Personal Values to Contextualized Personalization: Bayesian Personalized Value Alignment for LLMs arXiv cs.AI
Back to the Definition: Estimating Step-Level Advantages via Trajectory Graphs for Agentic Reinforcement Learning arXiv cs.AI
When Does Action Credit Need Updating? arXiv cs.AI
AlphaDiverse: Post-Training Local Quantitative Research Agents for Diverse Exploration in Alpha Factor Mining arXiv cs.AI
MeshHeal: Two-Timescale Self-Healing for Gray Failures in Decentralized LLM Agent Networks arXiv cs.AI
SLCA-GRPO: Resolving Cross-Segment Credit Misattribution in Tool-Calling RL arXiv cs.AI
From Self-Distillation to Self-Practice: Privileged Information for Multi-Turn Agents arXiv cs.AI