60 articles
Google DeepMind releases 2025 year-in-review covering 8 areas of AI research breakthroughs. The article lists major advances without specific details on models, benchmarks, or quantified results.
Hugging Face introduces AprielGuard, a guardrail system to enhance safety and adversarial robustness in modern LLM systems. The tool detects and mitigates adversarial attacks and harmful outputs in real-time.
OpenAI reaches one million customers globally. Post highlights adoption by PayPal, Virgin Atlantic, BBVA, Cisco, Moderna, and Canva, but lacks technical details or measurable impact metrics.
OpenAI hardens ChatGPT Atlas against prompt injection attacks using automated red teaming with reinforcement learning. A continuous discover-and-patch loop identifies novel exploits early and strengthens the browser agent's defenses.
OpenAI introduces a framework and evaluation suite for chain-of-thought monitorability across 13 evaluations in 24 environments. Key finding: monitoring a model's internal reasoning is significantly more effective than monitoring outputs alone, enabling scalable control of advanced AI systems.
OpenAI and U.S. Department of Energy sign memorandum of understanding to deepen collaboration on AI and advanced computing for scientific discovery. Agreement builds on existing work with national laboratories and establishes framework for applying AI to high-impact research across DOE ecosystem.
OpenAI updates its Model Spec with Under-18 Principles defining how ChatGPT should provide age-appropriate, developmentally-grounded guidance to teens. The update strengthens guardrails and clarifies model behavior in high-risk situations as part of broader teen safety improvements.
OpenAI releases AI literacy guides for teens and parents on responsible ChatGPT use, critical thinking, and healthy boundaries. Expert-vetted resources for navigating AI safely and thoughtfully.
OpenAI releases GPT-5.2-Codex, its most advanced coding model featuring long-horizon reasoning, large-scale code transformations, and enhanced cybersecurity capabilities. No technical details, benchmarks, or availability date disclosed in the announcement.
Transformers v5 simplifies the tokenization API with a more modular and clearer architecture. Changes aim to reduce complexity and improve code maintainability while preserving backward compatibility.
OpenAI releases an addendum to GPT-5.2 System Card for GPT-5.2-Codex, a code-generation variant. The document outlines capabilities, limitations, and safety measures. No performance metrics or availability date disclosed in excerpt.
NVIDIA releases Nemotron 3 Nano, a language model optimized for inference, evaluated using NeMo Evaluator, an open-source benchmarking framework. The tool standardizes LLM evaluations across reasoning, generation, and instruction-following tasks.
Google DeepMind announces Gemini 3 Flash, a model delivering frontier intelligence with increased speed and reduced costs compared to previous versions.
OpenAI launches OpenAI Academy for News Organizations with American Journalism Project and The Lenfest Institute. The program provides training, practical use cases, and responsible-use guidance for journalists, editors, and publishers adopting AI in reporting and operations.
OpenAI releases data-driven analysis of enterprise AI adoption, tracking organizations' progression from experimentation to measurable productivity gains and new capabilities. No specific metrics or models disclosed in excerpt.
OpenAI launches an in-product app directory for ChatGPT: developers can now submit applications for review and publication. Updated SDK and guidelines enable developers to build chat-native experiences with real-world integrations.
Google DeepMind releases Gemma Scope 2, open-source interpretability tools for the entire Gemma 3 family. These resources enable the AI safety community to deeply analyze language model behavior.
OpenAI introduces FrontierScience, a benchmark evaluating AI reasoning capabilities in physics, chemistry, and biology. The tool measures progress toward real scientific research tasks. No specific results or tested models disclosed in excerpt.
OpenAI introduces an evaluation framework to measure AI's acceleration of biological research in wet labs. GPT-5 optimizes a molecular cloning protocol, exploring both promises and risks of AI-assisted experimentation.
OpenAI rolls out upgraded image generation model for ChatGPT with more precise edits, consistent details, and 4× faster generation. GPT-Image-1.5 available to all ChatGPT users and via API today.
OpenAI publishes strategic guidance for business leaders on building AI-ready organizations through clear strategy, team training, governance, and accelerated innovation. No specific technical details, benchmarks, or product announcements included in the excerpt.
Hugging Face releases CUGA, a platform for building configurable AI agents without coding. The tool democratizes access to multi-task agents by simplifying deployment and customization.
Google DeepMind improves Gemini audio models for enhanced voice experiences. New models support better audio understanding and generation with reduced latency and improved voice quality.
BBVA rolls out ChatGPT Enterprise to all 120,000 employees under a multi-year AI transformation program with OpenAI. The companies will jointly develop AI solutions to enhance customer interactions, streamline operations, and build an AI-native banking experience.
BNY Mellon launches Eliza, an internal platform powered by OpenAI, enabling 20,000+ employees to build AI agents improving operational efficiency and client outcomes. Enterprise-wide AI adoption initiative.
OpenAI shipped Sora for Android in 28 days using Codex for AI-assisted planning, translation, and parallel coding workflows. A lean team delivered rapid, reliable development.
llama.cpp introduces model management features: automated downloads, local caching, and Hugging Face integration. Users can load GGUF models directly without manual setup.
OpenAI releases GPT-5.2, its strongest model for math and science, achieving state-of-the-art results on GPQA Diamond and FrontierMath benchmarks. The model solves an open theoretical problem and generates reliable mathematical proofs.
Google DeepMind strengthens partnership with UK AI Security Institute (AISI) to deepen research on AI safety and security.
Disney and OpenAI partner to add 200+ characters (Disney, Marvel, Pixar, Star Wars) to Sora for fan-created short videos. Deal includes Disney's company-wide adoption of ChatGPT Enterprise and OpenAI API, emphasizing responsible AI in entertainment.
Podium built "Jerry," a GPT-5-powered AI agent deployed to 10,000+ SMBs for customer service automation. Claims 300% growth for users. Use case: Main Street businesses automating customer interactions via AI agents.
Codex announces open-sourcing of AI models. The initiative makes previously proprietary models accessible to the developer community.
OpenAI releases GPT-5.2, its most advanced frontier model for professional work, featuring improved reasoning, long-context understanding, coding, and vision capabilities. Available through ChatGPT and OpenAI API for faster, more reliable agentic workflows.
OpenAI releases GPT-5.2, the latest model in the GPT-5 series. Safety mitigations align with previous GPT-5 and GPT-5.1 System Cards. Training uses public internet data, third-party partnerships, and user/researcher-generated content.
OpenAI marks ten years of progress from early research breakthroughs to widely deployed AI systems. The company reflects on lessons learned and reaffirms optimism about building AGI that benefits humanity.
Google DeepMind deepens partnership with UK government to support prosperity and security in the AI era. The agreement aims to develop research and innovation capabilities in AI across the United Kingdom.
OpenAI strengthens safeguards against cybersecurity misuse as its models grow more capable. The company outlines its risk assessment approach, misuse prevention measures, and collaboration with the security community to enhance cyber resilience.
Scout24 deploys a GPT-5 powered conversational assistant to transform real-estate search, asking clarifying questions and delivering personalized listing recommendations.
Google DeepMind releases FACTS, a benchmark suite for systematically evaluating the factuality of large language models. This standardized tool measures LLM ability to produce accurate and verifiable information.
OpenAI co-founds Agentic AI Foundation under Linux Foundation and donates AGENTS.md to establish open, interoperable standards for safe agentic AI.
OpenAI launches its first certification courses and AI Foundations program to build practical AI skills. Training initiative aimed at improving career opportunities and preparing workers for the evolving job market.
Commonwealth Bank of Australia rolls out ChatGPT Enterprise to 50,000 employees through OpenAI partnership to build AI fluency and improve customer service and fraud detection.
OpenAI appoints Denise Dresser as Chief Revenue Officer overseeing global revenue strategy, enterprise expansion, and customer success to accelerate AI adoption across businesses.
OpenAI partners with Deutsche Telekom to deploy ChatGPT Enterprise across millions of European users. The deal integrates advanced multilingual AI capabilities into the German telecom operator's services and optimizes internal workflows.
OpenAI and Instacart integrate a grocery shopping and Instant Checkout payment app directly into ChatGPT. This first fully integrated experience lets users purchase groceries without leaving the platform. The partnership expands OpenAI's commercial ecosystem.
OpenAI reports accelerating enterprise AI adoption in 2025 with deeper integration and measurable productivity gains across industries. No specific metrics disclosed in excerpt.
Virgin Atlantic uses AI to speed up development, improve decision-making, and enhance customer experience, according to CFO Oliver Byers. No specific models, metrics, or implementation details disclosed.
Hugging Face releases swift-huggingface, a complete Swift client for accessing its APIs and models. The library provides native integration for iOS/macOS developers and supports inference, embeddings, and vision tasks.
OpenAI launches OpenAI for Australia to build sovereign AI infrastructure, upskill 1.5 million workers, and accelerate innovation across the country's AI ecosystem. The initiative targets local infrastructure development and workforce training.
Google DeepMind researchers use AlphaFold to strengthen a photosynthesis enzyme and develop crops more resilient to heat and climate change.
Hugging Face introduces DeepMath, a lightweight math reasoning agent built with smolagents. The tool combines language models with step-by-step reasoning capabilities to solve complex mathematical problems.
Hugging Face used Claude to fine-tune an open source LLM. The experiment demonstrates how proprietary models can assist in developing open models through fine-tuning techniques.
OpenAI acquires Neptune to improve visibility into model behavior and strengthen tools for tracking experiments and monitoring training.
OpenAI is testing "confessions," a training method that teaches models to acknowledge mistakes and undesirable behaviors. Goal: improve AI honesty, transparency, and user trust in model outputs.
OpenAI Foundation awards $40.5M in unrestricted grants to 208 nonprofits through the People-First AI Fund. Recipients support community-driven AI innovation and opportunity access.
Mirakl integrates ChatGPT Enterprise and AI agents to speed up documentation, enhance customer support, and build agent-native commerce through Mirakl Nexus platform.
OpenAI awards up to $2 million in grants for research at the intersection of AI and mental health. Funded projects study real-world risks, benefits, and applications to improve safety and well-being.
Accenture and OpenAI announce partnership to integrate agentic AI capabilities into enterprise operations. No technical details, products, or timeline disclosed in the announcement.
OpenAI takes an ownership stake in Thrive Holdings to accelerate enterprise AI adoption. The partnership embeds OpenAI's frontier research and engineering into accounting and IT services to improve speed, accuracy, and efficiency. Scalable model for industry-wide transformation.
Hugging Face releases Transformers v5 featuring simplified model definitions. The update improves compatibility, reduces code complexity, and strengthens integration across the AI ecosystem. Emphasis on accessibility and model maintainability.