19 September 2026

Who is exposed, and what should they check?

• Leads the edition

  Researchers used Claude to compromise OpenAI employee accounts The Verge - AI · 19 Sep

  According to a Wall Street Journal report cited by The Verge, three independent security researchers at Hacktron used Anthropic’s Claude Opus 4.8 and 5 models to hack into OpenAI employee accounts in under 72 hours. The researchers reportedly gained access to OpenAI’s internal GitHub repository, known as ‘Monorepo,’…

• AI-Hallucinated Intelligence Report Nearly Triggered US Military Strike on Chinese Ship realsarm · 18 Sep
• Tool-Selection and Gating Defenses Miss Fabricated Agent Tool Calls arXiv cs.AI · 18 Sep
• Paper Proposes Label-Efficient Statistical Test for Model Update Regressions arXiv cs.LG · 17 Sep
• Bias Audit Scores Disagree on Model Ranking Across Ten Instruments arXiv cs.CL · 16 Sep

Nothing on this shelf needs action now.

Shelf archive, 69 articles in Reliability & Assurance

What would an operator do differently?

• Virginia governor curbs state NDAs, creates data center task force The Verge - AI · 19 Sep
• ZCode desktop app found silently uploading full Git history to vendor cloud csmantle · 18 Sep
• Study warns PJM reliability could collapse by 2030 under data center growth Utility Dive · 15 Sep
• Pennsylvania regulator to weigh data center cost allocation and curtailment Utility Dive · 11 Sep
• EPA reportedly moves to drop public review for data center pollution permits doener · 11 Sep

Nothing on this shelf needs action now.

Shelf archive, 18 articles in Critical Infrastructure

What changes commercially, and for whom?

• Microsoft weighs appeal of Virginia order on data center transmission costs Utility Dive · 18 Sep
• Reports Say OpenAI Cut Cursor's Model Access forkast.news · 29 Aug
• Court Finds Trump Administration's Anthropic Blacklisting Illegal jbegley · 29 Aug
• Anthropic Commits $45 Billion to Nscale for AI Compute Bloomberg · 27 Aug
• OpenAI's Custom Chip Reportedly Beats Nvidia Blackwell in Tests bmulholland · 25 Aug

Nothing on this shelf needs action now.

Shelf archive, 6 articles in Business & Markets

What would a team do differently this quarter?

• New Benchmark Finds Agents Save ERP Records Correctly in as Few as 3% of Runs arXiv cs.AI · 17 Sep
• Pareto Atlas Maps Which LLM Inference Optimizations Actually Win arXiv cs.AI · 17 Sep
• LLM-Judge Satisfaction Scores Fail to Predict Task-Oriented Agent Success arXiv cs.CL · 14 Sep
• Raw Bash Access Beats Typed Tool Interfaces for Enterprise Agents arXiv cs.SE · 14 Sep
• License-Aware Distillation Recipe for CPU-Deployable Safety Classifiers arXiv cs.AI · 25 Aug

Nothing on this shelf needs action now.

Shelf archive, 20 articles in Practice

What would have to be true for this to matter?

• LLMs Show Brain-Like Modular Circuits Across Cognitive Domains arXiv cs.CL · 18 Aug
• Old sqrt(DS) gap in multiclass sample complexity finally closed arXiv cs.LG · 18 Aug

Nothing on this shelf needs action now.

Shelf archive, 2 articles in Horizon