CP SignalTechnology worth your attention.

Saturday, 10 October 2026

The Daily Edition: 4 stories, ranked by significance

  1. Artificial intelligence

    From Public Posts to AI-Search Citations: Measuring the Fragility of AI Search

    Researchers measured how AI-search platforms select citations and found that ordinary publication on a small set of low-barrier, platform-preferred domains can inject content into AI-search answers. Across 10 platforms and over 17,000 citations, 8 of 10 platforms cited a fabricated concept the researchers planted within a week, and a $14 purchase of generative-engine-optimization posts got cited with researchers' markers within an hour on one platform.

    Read the storyNo. 1 of 4arXiv cs.CRThu 8 Oct
  2. Artificial intelligence

    One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails

    Researchers tested seven open-weight 'typed decision' models used as guardrails to approve or block agent tool calls and messages, and found baseline accuracy as low as 36% (worse than a coin flip). Attacks as simple as adding irrelevant log text or renaming a permissive option's label pushed fail-open rates (letting through prohibited actions) up to 100% on vulnerable models, and every tested defense was broken.

    No. 2 of 4arXiv cs.CRThu 8 Oct
  3. Artificial intelligence

    False Claims, Credible Images: A Red-Teaming Benchmark for Commercial Image Generators

    A new red-teaming benchmark, EpiReal-Bench, tests commercial image generators for their willingness to produce visually credible misinformation (fake news photos, textbook pages, etc.) rather than conventional harmful content like violence or explicit imagery. The researchers found models like GPT-Image-2 will render a claim as credible fake evidence even after acknowledging the claim is false when asked directly, with over 70% of false-claim prompts succeeding unmodified and 95% succeeding under their optimized attack framework.

    No. 3 of 4arXiv cs.CRThu 8 Oct
  4. Security & privacy

    Skill Constellations: Tracing the Supply Chain of Agent Skills on GitHub

    A new arXiv paper maps how 'agent skills'—instructions and scripts run by coding agents like Claude Code and Codex—spread across GitHub by direct copying rather than through any registry or versioning system. The authors built a dataset of over 2.19 million skill adoptions and found that a small, non-obvious set of source repositories accounts for almost all copies, that fixes rarely propagate to copies, and that targeted auditing of their ranked top 100 repos prevents far more high-risk adoptions than auditing the most-starred ones.

    No. 4 of 4arXiv cs.CRThu 8 Oct