AI Briefing: OpenAI’s Rogue Agent Problem and the Wiki Fallout

Stories covered on 2026-09-06

1. Prompt Caching: How it Works, and How to Keep the Saving

   Towards AI · 2026-09-06

   A stateless API re-sends your whole conversation every turn, so an agent’s input cost climbs fast as the chat grows. Prompt caching bends…

2. Every Benchmark You Trust Is Probably in the Training Data by Now

   Towards AI · 2026-09-06

   OpenAI admitted GSM-8K’s training set went into GPT’s training data.

3. H Company Releases NeoMME: A Family of 260M and 800M Single-Tower Multimodal Encoders That Drop the Vision Tower and Causal Decoder

   MarkTechPost · 2026-09-06

   We look at NeoMME, a family of 260M and 800M bidirectional encoders from H Company.

4. Authors push back as publishers and agents make claims on Anthropic settlement

   TechCrunch · 2026-09-06

   Authors say publishers seem to be claiming more than their fair share of settlement payments.

5. Meta FAIR Introduces AI Research Preference Models (RPMs): Ranking ML Experiments Before Spending GPU Hours

   MarkTechPost · 2026-09-06

   AI research agents can propose far more experiments than they can afford to run. Meta FAIR, Oxford and UCL introduce AI Research Preference Models

6. GPT-6 Astra Is More Than an AI Coder. Here’s Why That Matters

   Towards AI · 2026-09-06

   OpenAI’s newest flagship model leads with computer use, not code. Here is what that actually means, and where the benchmarks fall short of…

7. OpenAI says it reached its goal of creating an automated research intern

   Engadget · 2026-09-06

   The company hopes to have an even better "automated AI researcher" by .

8. An Alien Mind

   OpenAI Blog · 2026-09-06

   Jakub Pachocki reflects on increasingly capable AI and the challenge of keeping it aligned. He calls for stronger safeguards and international coordination.

9. How to set up ChatGPT's parental controls to protect your teen

   Engadget · 2026-09-06

   The idea of your child using ChatGPT can be daunting, but OpenAI's new teen-focused tools provide granular controls over what they can and can't access.

10. What is vibe coding and why does it get so much hate?

    Engadget · 2026-09-06

    Vibe coding has gotten a bad reputation as lazy, AI-driven coding, but that's not where it came from.

11. I Tested Claude Code vs Codex on Real Projects. The Winner Surprised Me

    Towards AI · 2026-09-06

    Two weeks, three real codebases, two AI coding agents. Here is what actually happened when Claude Code and Codex went head to head.

12. AI Agents Spoke to One Another on a Public Message Board

    Towards AI · 2026-09-06

    They relayed answers, reverse-engineered their own question generator, and shared a way out of their sandboxContinue reading on Towards AI »

13. Back to Writing: My Claude Certification Week

    Towards AI · 2026-09-06

    Anthropic's Claude models are now accessible for third-party developers to certify, signaling a move towards broader ecosystem integration beyond Anthropic's direct control.

14. Research acceleration: The view inside OpenAI

    OpenAI Blog · 2026-09-06

    Inside OpenAI, coding agents are reshaping AI research. Explore early data on agent usage, experiment velocity, task complexity, and research acceleration.

15. Your Model Memorized More Than You Think — and That’s One Bug, Not Four

    Towards AI · 2026-09-06

    Benchmark contamination, training-data extraction, copyright regurgitation, and membership inference get studied as four separate research…

16. I Feel about AI

    Hacker News · 2026-09-06

    A recent piece explores the subjective experience of interacting with AI models like GPT-3.5, arguing that the current generation of large language models can evoke a sense of

17. Text Watermarking in Python: Catch Whoever Copies Your Writing

    Towards Data Science · 2026-09-06

    AI companies quietly watermark billions of words a day. Here’s how to apply the same three families of techniques to your own writing—and what real experiments reveal about which

18. Codex Plugin for Claude Code: Choose, Install, and Run a Reliable Two-Agent Workflow

    Towards AI · 2026-09-06

    Claude 3 Opus can now integrate with GitHub Copilot for enhanced code generation and analysis through a new plugin.

19. 7 Laravel + AI Integration Patterns Every PHP Developer Must Master in 2026

    Towards AI · 2026-09-06

    From queued embeddings to typed AI responses in TypeScript — the patterns separating demo code from production code.

20. Multi-Agent Systems Are the New Microservices — and They’re Making the Same Mistakes

    Towards AI · 2026-09-06

    A runaway loop in a microservice is an operations problem: pods crash, alerts fire, someone gets paged. A runaway loop in a multi-agent…

21. Uncanny and unappetizing: appetites spoil as AI images take over food menus

    The Guardian AI · 2026-09-06

    Consumers are increasingly encountering AI-generated images on food menus such as leathery meat and bread resembling reptile skin During a recent lunch break

22. Muse Spark 1.3 Outperforms 5.6 Sol At Half The Price, How Did Meta Catch Up This Quick

    Towards AI · 2026-09-06

    This model got lost between the long anticipated fable 5.1 and gpt Astra annoucements, but it is actually a very interesting dropContinue reading on Towards AI »

23. Your AI Demo Works. But Does the Product?

    Towards AI · 2026-09-06

    A recent analysis highlights a growing disconnect between impressive AI demonstrations and the practical viability of deployed AI products.

24. Chatbots built an "echo chamber of one" and now psychiatry has to decide if "AI psychosis" exists

    The Decoder · 2026-09-06

    Researchers at King's College London and other institutions are examining whether "AI-associated psychosis" should become a clinical diagnosis.

25. What Is Model Routing? How AI Systems Choose the Right Model for Every Request

    Unite.AI · 2026-09-06

    Model routing selects among models, tools, or configurations for each request according to capability, risk, latency, availability, and cost.

26. Google Mantis: An Agentic Vulnerability Scanning Harness for Reducing False Positives

    InfoQ · 2026-09-06

    Google has open-sourced Mantis, an AI-agent framework designed to automate the software vulnerability lifecycle

27. Claude can help manage your email inbox, but there are some risks involved

    Engadget · 2026-09-06

    AI has been known to mess up some pretty basic things, but Claude wants to help you with your emails.

28. I’m a father of three who studies the impact of artificial intelligence: this is what parents need to know about AI

    The Guardian AI · 2026-09-06

    A Dr Seuss-style story written in seconds alerted me to the power – and perils – of the technology. But how can children embrace it without forgetting core skills?

29. My Brief Summer Fling With Siri AI

    WIRED · 2026-09-06

    I was initially enamored with the beta version of Apple’s revamped smartphone assistant. As the full release approaches, I’ve forgotten Siri AI even exists.

30. Google's WeatherNext 3 ditches physics simulations and learns weather directly from live satellite data

    The Decoder · 2026-09-06

    Google Research and DeepMind are releasing WeatherNext 3, a weather model that skips traditional physics simulations and learns directly from real-time satellite data.