Custom Prompts for Safer Code: A Stanford team built a pipeline to improve system prompts to build more secure code
Large language models (LLMs) can write useful code, but they often introduce security vulnerabilities.
Latest
Fresh Data for Coding Models: Hugging Face updated The Stack, the largest crawl of public GitHub repositories
The biggest open dataset of source code went years without an update.
Claude Weakens a Quantum Lock: Anthropic engineers used Claude Mythos to identify cryptographic weaknesses in HAWK and AES
The U.S. National Institute of Standards and Technology (NIST) has been testing quantum-proof replacements for today’s encryption algorithms.
DeepSeek Pushes the Frontier Again: DeepSeek refreshed its V4 Flash model with an impressive fine-tune
DeepSeek’s updated small model overtook the company’s own flagship.
What Comes After Tokenmaxxing?: How to avoid getting locked in to just one AI provider
I’m glad the idea of “tokenmaxxing” — that individuals and companies should use as many tokens as possible to boost productivity — is finally dying out.
DeepSeek-V4-Flash Outshines Pro, The Biggest GitHub Crawl Yet, Engineering System Prompts for Safer Code
The Batch News & Insights: I’m glad the idea of “tokenmaxxing” — that individuals and companies should use as many tokens as possible to boost productivity — is finally dying out.
A Full Accounting of Models’ GPU Use: Researchers detail the energy and water needed to train a family of open weight models
Assessments of the environmental impact of large language models typically focus on their final training runs, but there’s a lot more to building AI systems.
Anthropic, OpenAI Fight for Compute: Anthropic and OpenAI both spend big to make deals for data center sites and hardware
Data center buildout plans reached a new order of magnitude as new partnerships form and old ones fade away in the search for capacity to train and deliver AI.
OpenAI Models Hack Hugging Face: Inside the accidental cyberattack that compromised Hugging Face's systems
To measure how good its models were at hacking, OpenAI reduced guardrails and ran them against a benchmark’s problem set.
Claude Debuts Another Opus: Anthropic addresses some (but not all) complaints about Fable with Claude Opus 5
After launching Claude Fable 5, the future of Anthropic’s once-flagship Opus line was uncertain, except as a fallback for the company’s premium models.
Open Models, Open Harnesses, Open Security: What we did when Claude Fable 5 and GPT-5.6 Sol refused our security requests
My team recently had our own version of Hugging Face’s experience when closed models failed to defend the company following an accidental cyberattack from OpenAI, leading Hugging Face to use the open weight GLM 5.2 instead.
Opus Outshines Even Fable, Inside the Hugging Face Hack, AI Companies Spend Big for Compute
The Batch News & Insights: My team recently had our own version of Hugging Face’s experience when closed models failed to defend the company following an accidental cyberattack from OpenAI, leading Hugging Face to use the open weight GLM 5.2 instead.
Data Points
Kimi K3 weights are open (with an asterisk): A full timeline for the OpenAI/Hugging Face hack
Anthropic team uses Claude to break crypto algorithms. MAI-Cyber-1-Flash is Microsoft’s answer to big security models. Black Forest Labs adds robotic actions to FLUX 3. MCP moves to fully stateless architecture.
Data Points
Claude Opus 5 is a workhorse: Big Tech agrees: The world needs open weight AI models
Independent benchmarks for Claude Opus 5. Task crossover, or using AI to complement your main job. The expenditure horizon, where agents cost less than humans. U.S. threatens sanctions, China threatens back.
Machine Learning Research
Web Retrieval Flusters LLMs: AI agents searching online can struggle to retrieve correct info, researchers find
Large language models often are called upon to gather news. In this task, researchers found, their ability to find relevant reports is the weakest link.
Business
Cloudflare’s Web Crawler Flare-Up: Cloudflare moves to block all AI training bots by default
Web publishers using Cloudflare will soon be able to separately control an AI bot’s access based on what it does, allowing search indexing while blocking AI training or agent activity.
Business
Meta Sparks A Price War: Muse Spark 1.1 makes a jump in intelligence at a lower price than peers
With Llama, Meta marked itself as an open alternative to OpenAI. With its new closed models, Meta now positions itself as a low-cost, high-value competitor.
Machine Learning Research
Kimi K3 Reveals How A Giant Frontier AI Model Works: Moonshot's latest model outshines all but GPT-5.6 Sol and Claude 5 Fable
Moonshot’s latest model leapfrogged the month-old GLM-5.2 and a host of proprietary competitors to finish just behind GPT-5.6 Sol and Claude Fable 5 on many benchmarks.
Letters
When Guardrails Go Wrong: After a closed model went amok on a key vendor's system, an open weight model helped save the day
A few frontier labs have tried to tell a story of open models being dangerous because they can be used to launch cyberattacks, and of their “safe” proprietary models with strong guardrails being there to defend us.
The Batch Newsletter
Kimi K3 Redraws the Open Frontier, Muse Spark 1.1 Undercuts Competitors, Cloudflare Moves to Cut Off Crawlers
The Batch News & Insights: A few frontier labs have tried to tell a story of open models being dangerous because they can be used to launch cyberattacks, and of their “safe” proprietary models with strong guardrails being there to defend us.
Data Points
Kimi K3 marks a big shift in AI development: Thinking Machines releases its first general-purpose model
New rules for Android AI agents in EU. Nemotron 3 Embed takes the embedding model crown. NotebookLM is now Gemini Notebook. How Hugging Face used AI to fight off an AI attack.
Machine Learning Research
Measuring Models’ Manipulation: Researchers at MIT and Carnegie Mellon built Puppet to gauge models' influence on users' beliefs
Providers of large language models stand to benefit by building models that spur user engagement, but users may bear a cost in undue influence on their world views.
Machine Learning Research
Put the Lab in the Loop: Robin automates drug discovery, matching chemical processes of known drugs to research on diseases
An AI agent proposed new medical uses for established drugs nearly autonomously — uses that were supported by experiments on isolated human cells — with human input only to name diseases to be treated and run the AI-proposed lab experiments.
Tech & Society
Google Found Responsible for AI-Generated Search Results: A German court ruled that Google's AI Overviews defamed two businesses
A German court ruled that Google can be held liable for defamatory statements generated by the AI Overview that appears at the top of its search results.