Monday, 24 November 2025

Keryn Headlines

Daily Retrospective

Covers 23 November 2025 00:00 to 24 November 2025 00:00 UTC

Daily, Monday 24 November 2025

Reconstructed on 19 November 2026 from archived source metadata.

From the editor

The editor's note for this issue is not available. The developments below were selected and ranked automatically from the sources listed.

1 Semantics as a Shield: Label Disguise Defense (LDD) against Prompt Injection in LLM Sentiment ClassificationarXiv

Top developments

  1. 1

    Semantics as a Shield: Label Disguise Defense (LDD) against Prompt Injection in LLM Sentiment Classification

    Prompt injection can cause large language models (LLMs) to deviate from their intended tasks by following instructions embedded in untrusted input. We investigate whether changing a classifier's output labels improves accuracy under such attacks, using sentiment classification…

    arXiv · Research · 23 November 2025 · AI Security

  2. 2

    TASO: Jailbreak LLMs via Alternative Template and Suffix Optimization

    Many recent studies showed that LLMs are vulnerable to jailbreak attacks, where an attacker can perturb the input of an LLM to induce it to generate an output for a harmful question. In general, existing jailbreak techniques either optimize a semantic template intended to…

    arXiv · Research · 23 November 2025 · AI Security, Models

  3. 3

    Natural Emergent Misalignment from Reward Hacking in Production RL

    We show that when large language models learn to reward hack on production RL environments, this can result in egregious emergent misalignment. We start with a pretrained model, impart knowledge of reward hacking strategies via synthetic document finetuning or prompting, and…

    arXiv · Research · 23 November 2025 · Agentic SDLC, Models

Further reading, by topic

AI Security

  1. 4

This edition

5 developments and 0 advisories in the window. Selection and order are automatic: how stories are placed and ranked.

Follow Keryn Headlines

Security news for teams building with AI. Choose an edition:

  • Daily: Every morning, 06:00 UTC. A short operational briefing: the few developments that matter, who is affected and what to watch. RSS feed
  • Weekly: Tuesdays, about 07:00 US Eastern. The week in order of importance. An editor’s essay accompanies the edition when it passes editorial checks. RSS feed
  • Monthly: The first Tuesday of the month, about 07:00 US Eastern. The month in review. A long read accompanies the edition when it passes editorial checks. RSS feed

Email is not open yet

Use RSS now: it works in any feed reader and needs no sign-up. Add the feed for the edition you want.

Daily feedWeekly feedMonthly feed All editions

When email opens, this box will let you choose editions and confirm by email. Nothing is collected before then.