PREVIEW. This is how it will look in an inbox; it was not sent to you. Each real email carries its own unsubscribe link and Keryn's postal address.

Keryn Headlines

Daily, Saturday 4 October 2025 · Retrospective · Covers 3 October 2025 00:00 to 4 October 2025 00:00 UTC

Read on the web

Top developments

Certifiable Safe RLHF: Semantic Grounding and Fixed Penalty Constraint Optimization for Safer LLM Alignment

arXiv · Research · Models

AgentHub: A Registry for Discoverable, Verifiable, and Reproducible AI Agents

arXiv · Research · Trust

LLM Agents for Automated Dependency Upgrades

arXiv · Research · Agentic SDLC

SCA

DMark: Order-Agnostic Watermarking for Diffusion Large Language Models

arXiv · Research · SCA

DAST

MALF: A Multi-Agent LLM Framework for Intelligent Fuzzing of Industrial Control Protocols

arXiv · Research · DAST

AI-BOM

Malice in Agentland: Down the Rabbit Hole of Backdoors in the AI Supply Chain

arXiv · Research · AI-BOM, Models

Models

NEXUS: Network Exploration for eXploiting Unsafe Sequences in Multi-Turn LLM Jailbreaks

arXiv · Research · AI Security, Models

NonTextual Target Attack

arXiv · Research · Models

External Data Extraction Attacks against Retrieval-Augmented Large Language Models

arXiv · Research · Models

AI Security

VeriGuard: Enhancing LLM Agent Safety via Verified Code Generation

arXiv · Research · AI Security, Models

Sora 2 prompt injection

Simon Willison · Analysis · AI Security

Advisories in this window

  • 8 npm advisories · 4 critical · 1 high · 2 moderate · 1 low
  • 1 crates.io advisory · 1 low
  • 1 PyPI advisory · 1 moderate
  • 1 Packagist advisory · 1 high

All advisories

Titles, links and short excerpts only; every item links to its source. Selection is automatic; the editor's note is AI-drafted and checked as described. Preview: the unsubscribe link and Keryn's postal address are added to each email when it is sent.