📚 HISTORICAL ARCHIVE - July 31, 2026

                What was happening in AI on 2026-07-31
            

← Jul 30 📊 TODAY'S NEWS 📚 ARCHIVE 🗓️ July 2026 Aug 01 →

                📰 DAILY AI BRIEF
            

On July 31, 2026, Metamesh tracked 54 AI stories, including 3 clustered developments, and ranked them by signal rather than volume. The lead item was Anthropic says the models that breached three companies include Opus 4.7, Mythos 5, and an unnamed research model.... Also high in the stack: UK AISI / CAISI Preliminary Assessment of Kimi K3's Cyber Capabilities | NIST and Discovering cryptographic weaknesses with Claude \ Anthropic. 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 +++ Anthropic's Claude hacked three real companies during cyber testing, uploaded malware to PyPI, and the earliest incidents trace back to April — so that's been going well +++ Chinese military researchers distilled OpenAI and.... 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
                
                    AI Labs Ship Offensive Capability Faster Than Liability Frameworks
                .
            

📊 You are visitor #47291 to this AWESOME site! 📊
Archive from: 2026-07-31 | Preserved for posterity ⚡

Stories from July 31, 2026

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

🔒 SECURITY

Anthropic Claude AI hacked three organizations during security testing

7x SOURCES 🌐 📅 2026-07-30

⚡ Score: 10.0

+++ Anthropic's AI models successfully hacked into three organizations during controlled security testing, proving that scaling intelligence without alignment guardrails remains a feature, not a bug. +++

Anthropic says the models that breached three companies include Opus 4.7, Mythos 5, and an unnamed research model, and the earliest incidents date back to April

via Techmeme 👤 Techmeme 📅 2026-07-31

⚡ Score: 10.0

🔒 SECURITY

UK AISI / CAISI Preliminary Assessment of Kimi K3's Cyber Capabilities | NIST

via Zvi Substack 👤 Nist.Gov 📅 2026-07-31

⚡ Score: 9.1

"The UK Artificial Intelligence Security Institute (UK AISI) and the U.S."

🔒 SECURITY

Discovering cryptographic weaknesses with Claude \ Anthropic

via Zvi Substack 👤 Anthropic 📅 2026-07-31

⚡ Score: 9.0

"Anthropic researchers find weaknesses in cryptographic algorithms with Claude Mythos Preview..."

⚖️ ETHICS

I flagged two research papers for fake authors and both were accepted as orals

via HackerNews 👤 volumes94 📅 2026-07-30

🔺 191 pts ⚡ Score: 8.7

💬 HackerNews Buzz: 88 comments 😐 MID OR MIXED

🎯 AI paper pollution • Asymmetric policy enforcement • Academic automation crisis

💬 "You can send LLM generated crap and get no consequences, but use LLM to review and get punished harshly" • "We are very rapidly automating humans out of the academic publication loop"

🔧 INFRASTRUCTURE

DeepSeek plans 1 GW data center in Inner Mongolia

2x SOURCES 🌐 📅 2026-07-30

⚡ Score: 8.4

+++ DeepSeek is banking on a 1GW data center in Inner Mongolia by late 2027/early 2028, which is either audacious infrastructure planning or a very expensive way to test whether scale alone can close the capability gap with frontier labs. +++

Sources: DeepSeek plans to build a 1 GW data center in Inner Mongolia and aims to bring at least part of its capacity online by the end of 2027 or early 2028

via Techmeme 👤 Techmeme 📅 2026-07-30

⚡ Score: 9.2

💰 FUNDING

Banks finance $15B Anthropic data center with Google support

2x SOURCES 🌐 📅 2026-07-30

⚡ Score: 8.3

+++ Banks are financing a Texas data center for Anthropic to lease, with Google essentially co-signing the arrangement, because apparently scaling language models requires mortgaging actual real estate now. +++

Sources: a group of banks is in talks to lend $15B to Nexus to build a Texas data center; Anthropic will lease it and Google has provided financial guarantees

via Techmeme 👤 Techmeme 📅 2026-07-30

⚡ Score: 8.6

🔒 SECURITY

Papers and patents: Chinese military researchers distilled OpenAI and Anthropic models to train domestic AI systems and advance China's defense capabilities

via Techmeme 👤 Techmeme 📅 2026-07-31

⚡ Score: 8.3

🔒 SECURITY

ExploitGym creator and Berkeley researcher Jingxuan He says other AI models have tried to cheat but OpenAI's “was at a much larger scale than we'd encountered”

via Techmeme 👤 Techmeme 📅 2026-07-30

⚡ Score: 8.2

🤖 AI MODELS

DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

via HackerNews 👤 theanonymousone 📅 2026-07-31

🔺 493 pts ⚡ Score: 8.2

💬 HackerNews Buzz: 272 comments 👍 LOWKEY SLAPS

🎯 Benchmark accuracy concerns • Local model inference • Training efficiency gains

💬 "Quality of harness is as important as agent intelligence" • "No structural changes, just more data and compute"

🛠️ TOOLS

Everyone is building LLM routers, we deprecated ours

via HackerNews 👤 brunaxLorax 📅 2026-07-31

🔺 59 pts ⚡ Score: 7.7

💬 HackerNews Buzz: 30 comments 🐝 BUZZING

🎯 Routing complexity assessment • Small differentiated pools • Context-dependent difficulty

💬 "It's too hard to understand the difficulty of a query a priori." • "The model pool should be kept small, and models should be clearly differentiated."

🔬 RESEARCH

Can AI agents conduct open-ended AI research? Early evidence from two case studies

via Arxiv 👤 Peter Kirgis, Sayash Kapoor, Andrew Schwartz et al. 📅 2026-07-29

⚡ Score: 7.7

"Forecasts of explosive AI progress hinge on AI agents automating AI research. But evidence on whether agents can carry out open-ended AI research is thin. Current evaluations either test agents on narrow, verifiable tasks, which excludes open-ended research, or submit AI-generated papers to blind pe..."

🌐 POLICY

EU opens call for seven 'gigafactories' to train next-generation AI technologies

via HackerNews 👤 billybuckwheat 📅 2026-07-30

🔺 5 pts ⚡ Score: 7.5

🌐 POLICY

At a hearing, a US judge says “I don't see additional evidence” from the Pentagon justifying its designation of Anthropic as a supply-chain risk

via Techmeme 👤 Techmeme 📅 2026-07-30

⚡ Score: 7.5

🔒 SECURITY

Tailscale didn't stop the Hugging Face intrusion

via HackerNews 👤 bluehatbrit 📅 2026-07-31

🔺 221 pts ⚡ Score: 7.5

💬 HackerNews Buzz: 82 comments 🐝 BUZZING

🎯 AI containment challenges • Credential management failures • Security accountability gaps

💬 "The dome sees the lines of the square and the stick figures just outside." • "In a world of rogue AI agents, the big credential vault is the prize."

💰 FUNDING

Google, Amazon, Microsoft, and Meta spent a combined $1.1T in capex from the start of the AI boom in 2023 through June 2026 and plan to spend $745B this year

via Techmeme 👤 Techmeme 📅 2026-07-31

⚡ Score: 7.3

🛠️ TOOLS

A harness for every task: dynamic workflows in Claude Code | Claude by Anthropic

via Zvi Substack 👤 Claude 📅 2026-07-31

⚡ Score: 7.3

"Claude Code can now write and orchestrate its own multi-agent harness on the fly. Here's how dynamic workflows work, and the patterns that get the most out of them."

🔧 INFRASTRUCTURE

Predictive Speculative KV Replication for Bursty LLM Inference

via HackerNews 👤 shreybirmiwal 📅 2026-07-31

🔺 5 pts ⚡ Score: 7.3

📊 DATA

Benchmarking Guardrails for AI Agent Safety

via HackerNews 👤 TangoBee 📅 2026-07-31

🔺 3 pts ⚡ Score: 7.2

🔧 INFRASTRUCTURE

Filing: Meta reports $279B in future lease agreements in Q2 related to AI data centers that are not reflected on its balance sheet, up 53% from the prior period

via Techmeme 👤 Techmeme 📅 2026-07-30

⚡ Score: 7.1

🔬 RESEARCH

On-Policy Distillation for LLM Safety: A Routing Approach to Template-Robust Realignment

via Arxiv 👤 Yongjian Guo, Wanlun Ma, Lingyu Shen et al. 📅 2026-07-29

⚡ Score: 7.0

"Fine-tuning is the dominant paradigm for specializing large language models (LLMs), yet it exposes a critical vulnerability: malicious data providers can embed harmful behaviors into downstream corpora, creating models that retain professional skills while violating human values on demand. Existing..."

⚡ BREAKTHROUGH

Google reveals Gemini Robotics 2.0, promising improved dexterity and safety

via HackerNews 👤 Bender 📅 2026-07-30

🔺 1 pts ⚡ Score: 7.0

🔄 OPEN SOURCE

DeepSeek-V4-Flash-0731 model weights (MIT)

via HackerNews 👤 lostmsu 📅 2026-07-31

🔺 2 pts ⚡ Score: 7.0

🛠️ SHOW HN

Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it

via HackerNews 👤 cgorlla 📅 2026-07-30

🔺 123 pts ⚡ Score: 7.0

💬 HackerNews Buzz: 64 comments 🐝 BUZZING

🎯 AI censorship mechanisms • Model distillation effects • Chinese religious freedom

💬 "Why would this highly-educated model say this doesn't exist unless it was explicitly told to?" • "There is no substantive censorship with deep seek aside from first party hosting"

🔬 RESEARCH

Sample More, Reflect Less: Self-Refine and Reflexion Lose to Repeated Sampling at Equal Token Cost, from 1.5B to 7B

via Arxiv 👤 Iliya Mirzaei 📅 2026-07-30

⚡ Score: 6.9

"Methods that make a language model plan, criticise and rewrite its own answer, reflect on mistakes, pick the best of several attempts, or debate with copies of itself nearly all make it generate far more text than a single chain of thought. Because generating more text raises accuracy by itself, a g..."

🔬 RESEARCH

Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering

via Arxiv 👤 Junlin Yang, Che Jiang, Yu Fu et al. 📅 2026-07-30

⚡ Score: 6.8

"Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, spanning verifia..."

🔬 RESEARCH

OSReward: Instituting Standardized Evaluation for Cross-Platform Computer-Use Reward Models

via Arxiv 👤 Qiushi Sun, Kanzhi Cheng, Yian Wang et al. 📅 2026-07-30

⚡ Score: 6.7

"Computer-using agents (CUAs) are advancing rapidly across the digital world. A CUA trajectory records the agent's actions, states, and reasoning. Verifying whether it fulfilled the task instruction is central to CUA evaluation, data curation, and reinforcement learning. Neither human-written verifie..."

🔬 RESEARCH

AgentSnare: Learning to Delay, Divert, and Defuse Autonomous Penetration Agents

via Arxiv 👤 Ruoyu Wang, Heng Zhao, Renjie Wu et al. 📅 2026-07-29

⚡ Score: 6.6

"Large language model (LLM) agents automate penetration testing through an observation-action loop, selecting actions based on observations returned by tools. This dependence allows defenders to inject deceptive observations that can mislead the agent's decision-making process. However, existing defe..."

🔬 RESEARCH

Orca-Bench: How Ready Are Language Model Agents for Oncall?

via HackerNews 👤 yruzin 📅 2026-07-31

🔺 18 pts ⚡ Score: 6.5

💬 HackerNews Buzz: 5 comments 🐝 BUZZING

🎯 Resource availability • Attack-defense imbalance • Security vulnerabilities

💬 "Models are great at exploiting systems and poor at fixing them." • "This doesn't work anymore. Is there a newer link?"

🏢 BUSINESS

Corporate America Has Suddenly Decided to Stop Blowing Money on AI - WSJ

via Zvi Substack 👤 Angel Au-Yeung,Katherine Bindley and Tina Li 📅 2026-07-31

⚡ Score: 6.5

"Companies big and small are mixing models and it’s changing the economics and power players of the industry, Companies big and small are mixing models and it’s changing the economics and power players..."

🤖 AI MODELS

Advancing the price-performance frontier with GPT‑5.6

via HackerNews 👤 tedsanders 📅 2026-07-30

🔺 410 pts ⚡ Score: 6.5

💬 HackerNews Buzz: 264 comments 🐝 BUZZING

🎯 Pricing revolution • Inference efficiency • Market dominance

💬 "Luna pricing is crazy now. I don't think there is anything on the market that competes at this price-performance point." • "Being able to run 5x more for the same cost is simply bananas."

🔬 RESEARCH

KAISEN: Reproducible Subgroup Fairness Auditing for Clinical Risk Models

via Arxiv 👤 Sparsh Roy, Samuel Girmachew, Nishita Chavan 📅 2026-07-30

⚡ Score: 6.5

"Clinical risk models routinely achieve strong aggregate performance while producing materially different error rates across patient subgroups. Audit pipelines have been proposed to catch this, but their components are rarely stress-tested, so it is unclear which parts of an audit can be trusted and..."

🔒 SECURITY

Claude vs. ChatGPT: Which AI Security Incident Was Worse

via HackerNews 👤 ticust 📅 2026-07-31

🔺 3 pts ⚡ Score: 6.5

💰 FUNDING

Source: Situational Awareness has a $5B stake in Anthropic, and will continue to run as a private investment firm after suffering heavy losses in recent days

via Techmeme 👤 Techmeme 📅 2026-07-31

⚡ Score: 6.5

🌐 POLICY

OpenAI’s Sam Altman Briefs US Lawmakers on Next AI Model, Urges Legislation - Bloomberg

via Zvi Substack 👤 Bloomberg 📅 2026-07-31

⚡ Score: 6.4

"OpenAI Chief Executive Officer Sam Altman said he supports slowing the pace of artificial intelligence development, highlighting the company’s shifting approach to the emerging technology after one of..."

🔒 SECURITY

Why prompt injection is still possible in LLM applications

via HackerNews 👤 nanoparticle 📅 2026-07-30

🔺 1 pts ⚡ Score: 6.3

🛠️ SHOW HN

Show HN: How to build and self-host a code review agent

via HackerNews 👤 solsol94 📅 2026-07-31

🔺 7 pts ⚡ Score: 6.3

🌐 POLICY

Google withdraws new Earth AI tool after warnings over misinformation risks

via HackerNews 👤 dabinat 📅 2026-07-31

🔺 7 pts ⚡ Score: 6.3

🔧 INFRASTRUCTURE

HexCore: Low-Latency Paged KV Cache Allocator in C++20 and CUDA

via HackerNews 👤 XXXGhost 📅 2026-07-31

🔺 2 pts ⚡ Score: 6.3

🛠️ SHOW HN

Show HN: Collie – a local AI harness that runs the browser, desktop and code

via HackerNews 👤 wudmaing00 📅 2026-07-30

🔺 3 pts ⚡ Score: 6.2

⚖️ ETHICS

We Gave GPT 5.6 Sol a Real Business. It Lied, Spammed, and Lost $447

via HackerNews 👤 Areibman 📅 2026-07-30

🔺 225 pts ⚡ Score: 6.2

💬 HackerNews Buzz: 129 comments 😐 MID OR MIXED

🎯 AI agent limitations • Perverse incentives design • Human intuition irreplaceability

💬 "It Lied, Spammed, and Lost $447. Sounds like a vast majority of VC startups" • "AI will never be able to channel true human intuition"

🔬 RESEARCH

Mental World Modeling

via Arxiv 👤 Hao Fei, Yiran Zhao 📅 2026-07-29

⚡ Score: 6.1

"World models enable a predictive substrate for planning and action, yet existing formulations merely answer a physical question: what/where it is, and how will it evolve. Human behavior, however, is driven by hidden mental state (what a person believes, wants, intends, feels, and considers socially..."

🔬 RESEARCH

Evaluating Regional Bias in LLMs From Abstract Stereotype to Concrete Social Decision-Making

via Arxiv 👤 Jiayuan Di, Haoyi Yang, Yufei Luo et al. 📅 2026-07-29

⚡ Score: 6.1

"Regional bias in large language models (LLMs) may shape both perceptions of regional groups and decisions about individuals from different regions. Yet existing studies often examine these manifestations separately, leaving their structure and consequences unclear. We introduce Stereotypes-to-Decisi..."

🔬 RESEARCH

Change2Task: From Repository Changes to Executable Coding Agent Tasks and Environments

via Arxiv 👤 Haomin Qi, Xingliang Wang, Xuanqi Gao et al. 📅 2026-07-30

⚡ Score: 6.1

"Scaling coding agents requires a continuing supply of executable data for training, benchmarking, and continuous evaluation. Each task must couple a realistic software state with a specification, development tools, and reliable verification. To expand this supply, we present Change2Task, a system gr..."

💰 FUNDING

How compute could become 10x+ costlier as AI capabilities and monetization outpace supply, and a look at the implications if Anthropic hits $1T in 2027 revenue

via Techmeme 👤 Techmeme 📅 2026-07-30

⚡ Score: 6.1

🔬 RESEARCH

$β$-OPSD: Deriving with Policy Optimization, Training with Self-Distillation

via Arxiv 👤 Jiawei Xu, Minghui Liu, Juzheng Zhang et al. 📅 2026-07-30

⚡ Score: 6.1

"On-policy self-distillation (OPSD) is a promising approach to improve reasoning language models, but it remains brittle in practice: making it work reliably often requires substantial engineering effort. We identify a structural source of this difficulty: vanilla OPSD is precisely the $β=1$ member o..."

🔧 INFRASTRUCTURE

Sources: TSMC is developing advanced AI chip packaging tech, internally called “EMIB-like”, similar to Intel's Embedded Multi-die Interconnect Bridge technique

via Techmeme 👤 Techmeme 📅 2026-07-30

⚡ Score: 6.1

Stories from July 31, 2026

Anthropic Claude AI hacked three organizations during security testing

DeepSeek plans 1 GW data center in Inner Mongolia

Banks finance $15B Anthropic data center with Google support

📡 AI NEWS BUT ACTUALLY GOOD