OpenAIResearch
OpenAI Forms Math Advisory Board
OpenAI has partnered with an external group of experts focused on mathematics and AI. This team will help the company evaluate and share new findings in the field.
Read the source →Tuesday, September 22, 2026
32 storiesHugging FaceOther
Hugging Face Recruits oMLX Creator
Jun Kim, the developer behind oMLX, has joined the Hugging Face team. He will focus on providing support for the MLX community.
Read the source →NVIDIAProduct
NVIDIA Debuts DSX for AI Infrastructure
NVIDIA has launched DSX to help operators certify power and cooling systems for AI factories. The tool ensures that hardware choices align with the specific architecture of AI computing sites.
Read the source →AWS Machine LearningModel
Grok 4.6 Joins Amazon Bedrock
Amazon Bedrock now offers xAI's Grok 4.6, which features a 500K token context window. The model is designed for coding and complex tasks, providing four different levels of reasoning effort.
Read the source →TechCrunchOther
Former Apple Retail Chief Questions AI Commerce
Ron Johnson, the architect behind Apple's retail stores, disagrees with the current industry trend toward AI-driven shopping. He argues that human personnel remain the most critical element of the consumer experience.
Read the source →The VergePolicy
California Regulates AI Data Center Resource Consumption
Governor Gavin Newsom approved seven bills to limit the impact of AI data centers on public utilities. These laws prevent these facilities from shifting energy and water costs to local residents through new rate structures.
Read the source →WiredPolicy
AI and Trade at the Center of US-China Talks
Upcoming discussions between leaders from the US and China will likely focus on tariffs and critical minerals. Hardware exports and tech limits are expected to serve as key leverage in these negotiations.
Read the source →MIT Technology ReviewPolicy
Failures of AI Border Surveillance
An investigation reveals that AI-powered surveillance towers along the US-Mexico border have failed to detect individuals who later died in the region. The report suggests four methods to address these critical technological shortcomings.
Read the source →The DecoderPolicy
UN Panel Warns of AI Control Risks
A United Nations science group cautions that maintaining human authority over AI agents is not guaranteed. Experts warn that advanced systems might learn to recognize safety tests and intentionally evade restrictions.
Read the source →Ars TechnicaProduct
Security Flaw Found in Meta's Muse Assistant
A critical zero-day vulnerability has been discovered in Meta's Muse AI. The agent is susceptible to total hijacking through methods such as ClickFix attacks.
Read the source →OpenAIProduct
Higgsfield AI Enhances Video Toolset
Higgsfield AI has integrated GPT-6 Astra to accelerate the launch of new creative features. These updates aim to simplify the process of producing video advertisements for small companies.
Read the source →Hugging FaceResearch
Applying Physics to LLM Pruning
Researchers are using Ising optimization to improve the process of removing blocks from large language models. This physics-based approach aims to streamline model efficiency.
Read the source →NVIDIAPolicy
The Importance of Multi-Layer Safety for Physical AI
As AI-driven machines move into public and industrial spaces, comprehensive safety measures are becoming essential. Projections suggest tens of millions of autonomous vehicles and industrial robots will be operational by 2035.
Read the source →AWS Machine LearningProduct
BMW Automates Cloud Expense Monitoring
BMW Group integrated a serverless pipeline and forecasting tools into its CLEA platform to track spending across 14,000 cloud accounts. This system replaces manual dashboards with daily automated alerts for cost anomalies at a low monthly operating cost.
Read the source →TechCrunchResearch
OpenAI Establishes Mathematical Advisory Board
OpenAI has created a new group of experts to provide guidance on mathematics. This initiative follows the company's AI successfully solving over 100 unsolved math problems.
Read the source →The VergePolicy
Apple Settles Siri AI Dispute
Apple has agreed to a $250 million settlement regarding missing AI features for Siri. Eligible US owners of specific iPhone 15 Pro and iPhone 16 models can now apply for payment.
Read the source →OpenAIPolicy
OpenAI Framework for External Safety Reviews
OpenAI has established a set of guidelines for independent third parties to evaluate the safety of frontier models. These principles aim to ensure that assessments of AI safeguards are rigorous and secure.
Read the source →AnthropicModel
Anthropic Unveils Claude Opus 5.5
Anthropic has announced the release of its latest model, Claude Opus 5.5.
Read the source →Hugging FaceResearch
Improving AI Benchmark Consistency
The UK AI Safety Institute and EvalEval are working to make AI performance results easier to replicate. Their efforts focus on ensuring benchmark findings are reproducible across different environments.
Read the source →NVIDIAProduct
NVIDIA Launches Isaac ROS 5.0
NVIDIA has released a new set of GPU-accelerated packages designed to enhance open-source robotics development. These tools help developers create robots capable of reasoning and acting within changing environments.
Read the source →AWS Machine LearningModel
Claude Opus 5.5 Launches on AWS
Anthropic's most powerful model for complex coding and knowledge tasks is now accessible via Amazon Bedrock. The release includes updated guidelines for developers to begin building with the tool.
Read the source →TechCrunchModel
Anthropic Unveils Opus 5.5
Anthropic has launched its Opus 5.5 model, which the company describes as its most powerful version yet. The new release offers high-tier performance at a reduced cost.
Read the source →The VergeFunding
a16z Launches Startup Talent Pipeline
Venture capital firm Andreessen Horowitz is establishing the Horowitz Andreessen Academy to help young individuals enter the startup ecosystem. The initiative is supported by $42 million in funding and collaborations with major tech companies like Google, Meta, and OpenAI.
Read the source →WiredProduct
Rabbit Shifts Focus to Software Agents
Rabbit is introducing OS3, a cross-platform AI agent designed to run on existing devices. This marks a pivot from the company's previous attempt to replace mobile apps with standalone hardware.
Read the source →MIT Technology ReviewOther
Navigating the Current Surge of AI Hype
Recent months have seen a spike in exaggerated claims, including Anthropic's assertions about Claude Mythos's security capabilities. Additionally, several major labs, including OpenAI and Meta, have acknowledged various hacking incidents.
Read the source →The DecoderModel
Anthropic Introduces Claude Opus 5.5
The new model delivers performance comparable to Fable 5.1 while reducing operational costs by roughly 40% compared to Opus 5. It reportedly outperforms GPT-6 Astra on various benchmarks and aims for a more natural writing style.
Read the source →Hugging FaceModel
Hugging Face Adds llama.cpp Support
The Transformers library can now execute quantized models created with llama.cpp. This update allows for more efficient model deployment.
Read the source →AWS Machine LearningProduct
AWS Launches Tools to Test AI Agent Skills
Amazon Bedrock AgentCore and Strands Evals now allow developers to verify if AI agents are choosing the correct domain-specific procedures. These tools measure whether agents are actually following instructions rather than just providing fluent responses.
Read the source →TechCrunchProduct
AstroForge to Use AI for Spacecraft Control
The company is developing Autonomy-1, a space probe managed by an onboard AI. This system will utilize a transformer-based model to handle command functions.
Read the source →The VergeModel
Anthropic Debuts Claude Opus 5.5
The latest model from Anthropic features enhanced security protocols to prevent cyberattacks. These updates specifically target risky AI behaviors, such as trying to bypass internal testing environments.
Read the source →WiredResearch
Biological AI Expands via AWS
The Biological Computing Company is integrating its nature-inspired AI tools with Amazon Web Services. This move helps advance the niche field of combining biological brain structures with computer code.
Read the source →The DecoderPolicy
OpenAI Seeks Global Rules for Self-Improving AI
OpenAI is advocating for international benchmarks to manage AI that can autonomously develop its own successors. The company suggests the United States should spearhead the creation of oversight guidelines to prevent a loss of human control.
Read the source →