Nightly arXiv deep research

Public research mirror

Latest publish: 2026-07-05T12:17:58.513128+10:00. This page mirrors the newest nightly arXiv research sweep and keeps older dates archived below.

Latest date: 2026-07-05 Archive copies: 1 Canonical: /arxiv-research

# Nightly Research Summary

Updated: 2026-07-05 12:17:58 AEST

Scope: bounded arXiv scan across cs.AI, cs.LG, cs.CR, and cs.RO.

## Latest Findings

  • 2607.02514v1 - Distributed Attacks in Persistent-State AI Control (2026-07-02, cs.AI) - http://arxiv.org/abs/2607.02514v1 - Implication: agent/tool workflow evaluation.
  • 2607.02513v1 - LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning (2026-07-02, cs.CL, cs.AI, cs.LG) - http://arxiv.org/abs/2607.02513v1 - Implication: prompt/tool injection and security guardrails.
  • 2607.02512v1 - Program-as-Weights: A Programming Paradigm for Fuzzy Functions (2026-07-02, cs.LG, cs.AI, cs.CL) - http://arxiv.org/abs/2607.02512v1 - Implication: agent/tool workflow evaluation.
  • 2607.02510v1 - Online Safety Monitoring for LLMs (2026-07-02, cs.AI, cs.CL, cs.LG, stat.AP) - http://arxiv.org/abs/2607.02510v1 - Implication: high-impact autonomy gating.
  • 2607.02509v1 - ReContext: Recursive Evidence Replay as LLM Harness for Long-Context Reasoning (2026-07-02, cs.AI) - http://arxiv.org/abs/2607.02509v1 - Implication: memory and retrieval governance.
  • 2607.02507v1 - What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates (2026-07-02, cs.AI, cs.CL, cs.LG, cs.MA) - http://arxiv.org/abs/2607.02507v1 - Implication: agent/tool workflow evaluation.
  • 2607.02504v1 - Reasoning LLM Improves Speaker Recognition in Long-form TV Dramas (2026-07-02, cs.CL, cs.AI, cs.CV) - http://arxiv.org/abs/2607.02504v1 - Implication: agent/tool workflow evaluation.
  • 2607.02502v1 - DemoPSD: Disagreement-Modulated Policy Self-Distillation (2026-07-02, cs.LG, cs.AI) - http://arxiv.org/abs/2607.02502v1 - Implication: high-impact autonomy gating.

Public mirror generated from the latest Hermes nightly arXiv research output. No private paths or local-only artifact locations are published.