Codex Security: now in research preview
OpenAI launches Codex Security in research preview: an AI agent that detects, validates, and patches application security vulnerabilities.
OpenAI News·6 Mar
Search signals, briefings, company results, benchmarks and glossary terms.
Search signals, briefings, company results, benchmarks and glossary terms.
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
OpenAI launches Codex Security in research preview: an AI agent that detects, validates, and patches application security vulnerabilities.
Balyasny Asset Management deployed OpenAI-powered agent workflows to automate and scale investment research processes.
Descript used OpenAI reasoning models to automate multilingual video dubbing, preserving timing and meaning at scale.
Hugging Face published research on optimizing Vision-Language Action (VLA) models for deployment on embedded robotics platforms.
OpenAI research shows that reasoning models struggle with 'chain-of-thought' control, highlighting the ongoing need for external monitoring.
OpenAI announces GPT-5.4, claiming top performance in coding, computer use, tool search, and 1M-token context window.
OpenAI published a system card for GPT-5.4 Thinking, a reasoning-focused model variant in its GPT-5 family.
OpenAI introduced new tools, certifications, and resources aimed at educational institutions to address AI capability gaps and expand learning opportunities.
OpenAI presented a framework of five AI value models, from workforce fluency to process reinvention, for enterprise AI adoption.
OpenAI launches ChatGPT integration for Excel and financial apps, powered by GPT-5.4, targeting regulated environment workflows.
A diverse coalition of conservative, progressive, and civil society groups released shared AI principles for a 'pro-human' movement.
Google DeepMind released Gemini 3.1 Flash-Lite, a faster and more cost-efficient version of its Gemini 3 series model.
OpenAI published a 'System Card' for an unreleased model, GPT-5.3 Instant, suggesting a future model family or a new product tier.
OpenAI released GPT-5.3 Instant, described as offering smoother, more useful everyday conversations.
OpenAI published details on a contract with the US Department of Defense, outlining safety guidelines and deployment in classified environments.
Google Chrome is developing quantum-safe HTTPS certificates and addressing performance challenges for quantum-resistant cryptography in TLS connections.
OpenAI and Amazon announced a strategic partnership to bring OpenAI's Frontier platform to AWS, focusing on infrastructure and custom models.
AWS Bedrock introduced a stateful runtime environment for agents, enabling persistent orchestration and memory for multi-step AI workflows.
OpenAI announces $110B funding round at $730B valuation, with $30B SoftBank, $30B NVIDIA, $50B Amazon.
OpenAI published updates on its mental health safety work, detailing parental controls, trusted contacts, distress detection, and litigation status.
OpenAI and Pacific Northwest National Laboratory partnered to create DraftNEPABench, evaluating AI coding agents for federal permitting, claiming 15% drafting time reduction.
OpenAI Codex integrates with Figma to enable bidirectional code-design workflows, aiming to accelerate product iteration.
Google AI prevents over 10 billion suspected malicious calls and messages monthly on Android, strengthening scam protection features.
OpenAI's Feb 2026 threat report details how bad actors use AI combined with web and social platforms, and outlines detection/defense responses.
OpenAI appointed Arvind KC as Chief People Officer to scale the company and evolve its work culture in the age of AI.
A new paper by AI Snake Oil quantifies the gap between AI agent capabilities and their real-world reliability, proposing a science for measurement.
Import AI #446 covers nuclear energy for AI, a Chinese AI benchmark, and AI measurement in policy contexts.
OpenAI launches Frontier Alliance Partners programme to help enterprises scale AI agents from pilot to production deployment.
GGML and llama.cpp, key projects for efficient local LLM inference, have joined Hugging Face to ensure their long-term development.
Hugging Face and Unsloth announced free fine-tuning of AI models, potentially reducing GPU costs and accelerating model development.
Google is leveraging AI to combat AI-driven fraud, malware, and privacy invasions within its Android and Google Play ecosystems in 2025.
Google DeepMind announced Gemini 3.1 Pro, a new model for complex tasks and longer context windows, building on the Gemini family.
OpenAI commits $7.5M to The Alignment Project, funding independent AI alignment research focused on AGI safety and security risks.
OpenAI announces India expansion: local infrastructure build-out, enterprise partnerships, and workforce upskilling programmes.
Google DeepMind's Gemini app now integrates Lyria 3, enabling users to generate 30-second music tracks from text or images.
One Useful Thing outlines a framework for categorizing and selecting AI systems beyond basic chatbots for 'agentic' applications.
OpenAI and Paradigm launch EVMbench to evaluate AI agents on detecting, patching, and exploiting smart contract vulnerabilities.
BIS Basel Committee published guidance on Synthetic Risk Transfers, a mechanism for banks to shed credit risk from asset pools.
Google DeepMind launches National Partnerships for AI in India, focusing on AI for science and education to accelerate discovery.
NIST announces the AI Agent Standards Initiative to develop interoperable and secure standards for the next generation of AI agents.
Bank of England held roundtables with regulated firms to understand constraints in AI/ML adoption.
OpenAI's GPT-5.2 derived a new theoretical physics formula for gluon amplitude, subsequently proved and verified by collaborators.
OpenAI adds Lockdown Mode and Elevated Risk labels to ChatGPT to counter prompt injection and AI-driven data exfiltration.
OpenAI releases GABRIEL, an open-source toolkit using GPT to convert qualitative text/images into quantitative data for social science research.
OpenAI describes its infrastructure for managing real-time access to Codex and Sora via rate limits, usage tracking, and credits.
Hugging Face released custom kernels derived from OpenAI Codex and Anthropic Claude for tailored model optimization.
Analysis suggests AI may not inherently reduce legal service costs, challenging claims of automatic efficiency gains in professional services.
Google DeepMind announced 'Gemini 3 Deep Think', an updated specialized reasoning mode for science, research, and engineering problems.
Research trends focus on improving AI model efficiency across ML, Robotics, CV, and NLP to reduce computational resource usage.
NIST allocated over $3 million to eight small businesses via the Small Business Innovation Research (SBIR) program for AI, biotech, semiconductors, and quantum.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion