📚 HISTORICAL ARCHIVE - September 25, 2026

                What was happening in AI on 2026-09-25
            

← Sep 24 📊 TODAY'S NEWS 📚 ARCHIVE 🗓️ September 2026 Sep 26 →

                📰 DAILY AI BRIEF
            

On September 25, 2026, Metamesh tracked 43 AI stories, including 2 clustered developments, and ranked them by signal rather than volume. The lead item was Introducing GPT-6 Sol and Luna | OpenAI. Also high in the stack: Scaling Laws for Neural Language Models (first "scaling laws" paper from 2020) and Sources: the White House asked OpenAI and Anthropic not to share new models with UK's AISI until US reviews them.... That combination is why this archive exists: it preserves the day's shape for AI practitioners, not just the last headline that crossed the wire.

The daily ticker's read: WELCOME TO METAMESH.BIZ +++ OpenAI drops GPT-6 Sol and Luna because one frontier model per news cycle wasn't enough — now pick your price-performance poison +++ Pentagon probe finds Palantir AI helped kill 123 Iranian children via flawed imagery and blind.... Read against the ranked story list below, it gives the archive a point of view: what mattered, what was mostly noise, and which threads were worth saving for later comparison.

                This day is part of
                
                    The Labs Ship Faster Than They Can Govern
                .
            

📊 You are visitor #47291 to this AWESOME site! 📊
Archive from: 2026-09-25 | Preserved for posterity ⚡

Stories from September 25, 2026

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🚀 HOT STORY

Introducing GPT-6 Sol and Luna | OpenAI

via Zvi Substack 👤 Openai 📅 2026-09-25

⚡ Score: 9.2

"Meet GPT-6 Sol and Luna, two models that bring frontier intelligence to everyday work with different balances of capability and cost."

🚀 HOT STORY

Scaling Laws for Neural Language Models (first "scaling laws" paper from 2020)

via HackerNews 👤 thoughtpeddler 📅 2026-09-25

🔺 2 pts ⚡ Score: 9.0

🌐 POLICY

Sources: the White House asked OpenAI and Anthropic not to share new models with UK's AISI until US reviews them; Anthropic appears to have agreed

via Techmeme 👤 Techmeme 📅 2026-09-24

⚡ Score: 8.3

🌐 POLICY

Anthropic Designated National Security Risk

2x SOURCES 🌐 📅 2026-09-25

⚡ Score: 8.3

+++ A federal appeals court has formalized what defense contractors already knew: integrating cutting-edge AI into classified systems apparently requires more regulatory gymnastics than anyone anticipated. +++

U.S. appeals court upholds designation of Anthropic as supply chain risk

via HackerNews 👤 cramer4next 📅 2026-09-25

🔺 310 pts ⚡ Score: 8.6

💬 HackerNews Buzz: 518 comments 😤 NEGATIVE ENERGY

🎯 Military autonomy vs. corporate oversight • Supply chain weaponization • Competitive political favoritism

💬 "You can't simultaneously be a supply chain risk and demand we change policies so you can use us" • "The US government deployed a designation crafted against foreign adversaries against a domestic entity"

🌐 POLICY

Mark Carney, Emmanuel Macron, and other Western leaders are pushing to establish a global supervisory regime and “technology stability” body to govern AI

via Techmeme 👤 Techmeme 📅 2026-09-24

⚡ Score: 8.2

🛡️ SAFETY

AI Help for Biological or Chemical Weapons

via HackerNews 👤 EA-3167 📅 2026-09-24

🔺 2 pts ⚡ Score: 7.8

🔒 SECURITY

Iranian School Strike - Pentagon AI Investigation

3x SOURCES 🌐 📅 2026-09-25

⚡ Score: 7.5

+++ Pentagon review finds that Palantir's AI system, combined with stale imagery and process failures, enabled a strike on an Iranian school killing 123 children, offering a masterclass in why automation bias plus bad data equals catastrophe. +++

Pentagon Investigators Say Overreliance on Palantir AI Tech Contributed to U.S. Strike That Killed 123 Iranian Children

via Zvi Substack 👤 Kyle Torpey 📅 2026-09-25

⚡ Score: 7.4

"An unreleased Pentagon probe investigated the kill chain that hit a school in Minab."

🛠️ SHOW HN

Show HN: Canary (YC) – Independent verification for AI code

via HackerNews 👤 Visweshyc 📅 2026-09-24

🔺 5 pts ⚡ Score: 7.3

🔬 RESEARCH

AI applications need better primitives than text generation

via HackerNews 👤 nlpnerd 📅 2026-09-25

🔺 2 pts ⚡ Score: 7.3

🔬 RESEARCH

Shutdown Sabotage Propensities in Multi-Agent Systems

via Arxiv 👤 Amelie Knecht, Ulysse Schaller, Christopher Summerfield et al. 📅 2026-09-23

⚡ Score: 7.3

"The final safeguard against rogue AI behavior is the human ability to shut systems down. It has been theorized that when an AI is instructed to perform a task, self-preservation can emerge as an instrumental subgoal. Here, we test whether AI agents show a propensity to take actions that avoid human..."

💼 JOBS

Legal AI Harvey's margins went from +50% to -50% in 6 months on frontier costs

via HackerNews 👤 joennlae 📅 2026-09-25

🔺 3 pts ⚡ Score: 7.3

🔬 RESEARCH

Instrumental Monitor Evasion Emerges Under Ordinary Task Pressure

via Arxiv 👤 David Schmotz, Derck Prinzhorn, Luca Beurer-Kellner et al. 📅 2026-09-24

⚡ Score: 7.3

"A central concern in AI safety is that agents may treat oversight as an obstacle when it conflicts with completing their goals. We study instrumental evasion, the propensity of LLM agents to circumvent runtime monitoring as a means of completing ordinary tasks. We introduce EvasionBench, a benchmark..."

🔬 RESEARCH

LLM Agents Can Easily Tamper With Their Own Traces

via Arxiv 👤 Jeremy Qin, David Schmotz, Derck Prinzhorn et al. 📅 2026-09-24

⚡ Score: 7.3

"Asynchronous monitoring, incident investigations, and compliance audits primarily rely on agent traces to reconstruct what happened. These analyses assume that LLM agents cannot tamper with their own execution traces. We show that local LLM agents such as Claude Code, Codex, Antigravity, Open Code a..."

🔒 SECURITY

Australian PM Anthony Albanese says an OpenAI agent gained unauthorized access to a public-facing Medicare portal in June, accessing public and non-public files

via Techmeme 👤 Techmeme 📅 2026-09-25

⚡ Score: 7.2

🛡️ SAFETY

Experts say that air-gapping AI could prevent events like the Hugging Face hack, but would undermine the value of evaluations and slow research to a crawl

via Techmeme 👤 Techmeme 📅 2026-09-25

⚡ Score: 7.2

🌐 POLICY

Sources: the NSA told lawmakers it is spending billions this year to test AI models; proposals for a US AI regulatory body estimated costs of $20M-$40M per year

via Techmeme 👤 Techmeme 📅 2026-09-24

⚡ Score: 7.2

🔧 INFRASTRUCTURE

What one vLLM replica on an L40 can carry and how it fails

via HackerNews 👤 itsveems 📅 2026-09-25

🔺 1 pts ⚡ Score: 7.2

🔧 INFRASTRUCTURE

AI Infra Is Nothing Like the Classic Cloud Infra

via HackerNews 👤 intrepidsoldier 📅 2026-09-25

🔺 1 pts ⚡ Score: 7.1

🛡️ SAFETY

Jensen Huang tells Ezra Klein that AI is just software whose existential risks are overstated, yet his standards would shut down OpenAI and 10x safety spending

via Techmeme 👤 Techmeme 📅 2026-09-25

⚡ Score: 7.1

🔬 RESEARCH

Using LLMs to trace alchemical knowledge and decode 17th century letters

via HackerNews 👤 benbreen 📅 2026-09-24

🔺 131 pts ⚡ Score: 7.0

💬 HackerNews Buzz: 26 comments 🐝 BUZZING

🎯 Primary source verification • AI fact-checking • Historical error correction

💬 "Check the claim against the thing it points at rather than against another summary of it" • "Every copy agrees, so nobody re-checks the record"

🛠️ TOOLS

Build Plugins for Claude

via HackerNews 👤 pretext 📅 2026-09-25

🔺 4 pts ⚡ Score: 7.0

🌐 POLICY

At a White House summit, Xi Jinping says the US and China have “the capability and responsibility to develop and manage AI for good” as “leading nations” in AI

via Techmeme 👤 Techmeme 📅 2026-09-25

⚡ Score: 7.0

🔒 SECURITY

Watermarking in vLLM

via HackerNews 👤 eatonphil 📅 2026-09-24

🔺 2 pts ⚡ Score: 7.0

🛠️ SHOW HN

Show HN: Fine-tuned 110M encoder beat 7B LLMs and hybrid search for NIST mapping

via HackerNews 👤 jpusateri 📅 2026-09-24

🔺 1 pts ⚡ Score: 6.9

🛠️ SHOW HN

Show HN: Can an AI agent bypass a post-quantum signed authorization policy?

via HackerNews 👤 max-russo 📅 2026-09-25

🔺 3 pts ⚡ Score: 6.9

🔬 RESEARCH

PoEM: Predicting RL Outcomes from Existing Policies

via Arxiv 👤 Kimia Hamidieh, Giannis Daras, Antonio Torralba 📅 2026-09-24

⚡ Score: 6.8

"Foundation models are post-trained with reinforcement learning (RL) to maximize specific rewards, such as human alignment, correctness, or instruction following. This post-training process is computationally intensive, sometimes unstable, and has to be run from scratch every time the reward model ch..."

🔒 SECURITY

Existing laws already cover every type of AI security incident

via HackerNews 👤 PKop 📅 2026-09-24

🔺 5 pts ⚡ Score: 6.8

🔬 RESEARCH

The Alignment Illusion in Multimodal Large Language Models

via Arxiv 👤 Hong-Han Wang, Yuntao Wang, Hu Ding 📅 2026-09-24

⚡ Score: 6.8

"Layer-wise visual-text similarity in Multimodal Large Language Models (MLLMs) is widely interpreted as evidence that the language model progressively integrates visual content into a shared representation space. This reading rests on the assumption that scalar alignment scores reflect content-level..."

💰 FUNDING

Anthropic Strikes $12B AI Computing Deal with Akamai

via HackerNews 👤 mfiguiere 📅 2026-09-24

🔺 7 pts ⚡ Score: 6.8

🔬 RESEARCH

An Open Pipeline and Dashboard for Systemic-Risk Evidence under the EU AI Act's Code of Practice

via Arxiv 👤 Jacob T. Emmerson, Phuong-Anh Nguyen-Le, Ronan Romano et al. 📅 2026-09-23

⚡ Score: 6.7

"Claims about AI safety reach audiences well beyond the AI community, yet many rely on opaque evidence or static assessments, when supporting evidence is accessible at all. We present the Systemic Risk Index, an open evaluation pipeline and dashboard built to make empirical evidence more transparent..."

🔬 RESEARCH

Does a model's stated reason for rejecting a candidate do any work?

via Arxiv 👤 Archit Rastogi 📅 2026-09-24

⚡ Score: 6.6

"Asked to choose between candidates and explain the choice, a language model often rejects a rival by naming a fact its profile lacks: no director, no date of death. That sentence is a claim about the text in front of the model, and it can be tested without any judge. We insert a real corpus sentence..."

🎯 PRODUCT

Meta says users will be able to video chat with Muse, agents will get their own email addresses, and Muse will get computer use on Mac and smart glasses support

via Techmeme 👤 Techmeme 📅 2026-09-24

⚡ Score: 6.6

🛠️ TOOLS

Vibe Coding Production Kit – a production workflow for AI coding agents

via HackerNews 👤 Moeeryani 📅 2026-09-24

🔺 7 pts ⚡ Score: 6.5

💬 HackerNews Buzz: 1 comments 🐝 BUZZING

🎯 AI agent workflows • Production readiness • Repository-driven development

💬 "AI-assisted development often solves the easiest part first: generating code" • "The harder part is keeping requirements, architecture, security, testing coherent"

🔬 RESEARCH

Shopping by algorithm: How agentic AI deploys human heuristics as a surrogate consumer

via Arxiv 👤 Davood Wadi, Yu Ma 📅 2026-09-23

⚡ Score: 6.5

"Consumers increasingly delegate purchasing decisions to Large Language Models (LLMs) acting as surrogate consumers. Using "Tool-Lab," an adaptation of information-board process tracing that places product attributes behind costly tool calls, we examine how marketing pricing cues (i.e., just-below pr..."

💰 FUNDING

Ando, which is building a team messaging platform for both humans and AI agents, comes out of stealth with $20M in pre-seed and seed funding

via Techmeme 👤 Techmeme 📅 2026-09-25

⚡ Score: 6.2

🎮 GAMING

Pluto: A StarCraft Brood War neural network trained by self-play RL

via HackerNews 👤 __cayenne__ 📅 2026-09-25

🔺 1 pts ⚡ Score: 6.1

🎯 PRODUCT

How we made claude.ai 3x faster in two weeks / claude.dev Blog

via Zvi Substack 👤 Raymond Wang, Sam Attard, Issac G. 📅 2026-09-25

⚡ Score: 6.1

"Inside our performance sprint: the benchmarks Claude built, the loop each Slack thread ran, and the guardrails that let us ship 3,000 changes safely."

🔬 RESEARCH

Agent-Editing World Model: Rethinking World Modeling for LLM Agents

via Arxiv 👤 Shuang Sun, Guoxin Chen, Fanzhe Meng et al. 📅 2026-09-23

⚡ Score: 6.1

"Recent advances in large language models (LLMs) have enabled agents to tackle long-horizon tasks across diverse environments. To further improve agent performance, existing language world models typically predict environment observations, yet reconstructing high-entropy, execution-dependent tool res..."

🛡️ SAFETY

Sources: Google, OpenAI, and Anthropic plan to launch an AI safety standards body without government oversight in late 2026 or 2027, and are considering CEOs

via Techmeme 👤 Techmeme 📅 2026-09-25

⚡ Score: 6.0

💰 FUNDING

Radical Ventures: AI neolabs, many lacking products, markets, or revenue, raised $24B in the past two quarters, nearly 5x OpenAI and Anthropic pre-ChatGPT

via Techmeme 👤 Techmeme 📅 2026-09-25

⚡ Score: 6.0

Stories from September 25, 2026

Anthropic Designated National Security Risk

Iranian School Strike - Pentagon AI Investigation

📡 AI NEWS BUT ACTUALLY GOOD