Here is today's AI news. Story 1. OpenAI. OpenAI launched GPT-6 Astra, a new flagship model emphasizing cybersecurity and computer-use capabilities, and also committed $1 billion in subsidized access to AI cyberdefense tools amid rising AI security concerns.. OpenAI launched GPT-6 Astra, its most capable model yet, with advanced computer-use, coding and cybersecurity abilities. The model is the first to reach OpenAI’s “Critical” cyber-capability threshold, prompting added monitoring that may pause or stop tasks. OpenAI also pledged $1 billion in subsidized access, training and support through Daybreak for Frontline Defenders, initially targeting U.S. utilities, governments, community banks and nonprofits. Story 2. Google. Google DeepMind released Gemini 3.8 Flash, along with a cyber-focused variant, highlighting improved coding and agentic-task performance.. Google DeepMind released Gemini 3.8 Flash and a restricted Gemini 3.8 Flash Cyber variant, aimed at agentic workflows, coding, and vulnerability discovery. Google says Flash improves software engineering and multi-step reasoning at 3.7 Flash’s speed and price, while Cyber is gated through the Fairwind program for trusted defenders. It underscores AI’s shift toward cheaper, more capable developer and security tools. Story 3. Meta. Meta unveiled Muse Spark 1.3, a new model line update aimed at better efficiency in coding and agentic workloads.. Meta unveiled Muse Spark 1.3 on Sept. 2, 2026, calling it the company’s biggest improvement yet for coding and agentic tasks. The model launched in Muse Code and the Meta Model API, with rollout planned for Meta AI, Facebook and Instagram. Meta says it sustains longer workflows, asks clarifying questions and uses about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2. The release intensifies competition with Anthropic and OpenAI, while a higher-reasoning version remains under safety testing. Story 4. Anthropic. Anthropic shipped Claude Fable 5.1 and Claude Mythos 5.1, touting them as major upgrades for coding and knowledge work while also cutting cache-read pricing.. Anthropic released Claude Fable 5.1 and restricted-access Mythos 5.1, the same underlying model with different safeguards. Fable targets long-running coding, research and knowledge work, with a 1M-token context window and 128K-token outputs. Input/output remain $10/$50 per million tokens, while cache reads fell 75% to $0.25—potentially lowering costs for agentic enterprise workflows. Story 5. Microsoft. Microsoft AI released MAI-Transcribe-2, a speech model focused on voice agents, with reported 2.0% word error rate and very fast real-time performance.. Microsoft AI released MAI-Transcribe-2, a new speech-to-text model aimed at voice agents. Microsoft says it supports 60 languages and offers speaker diarization, word timestamps, language ID, and keyword biasing. Industry watchers say its ~2.0% word error rate and ~411x real-time speed make it a strong low-cost option for fast, accurate transcription. That's all for today's AI news briefing.