From The New Stack · 40 stories
AWS Continuum Integrates AI-Powered Penetration Testing into CI/CD Pipelines
AWS Continuum, formerly AWS Security Agent, now provides AI-driven penetration testing that integrates directly into CI/CD pipelines. This service automatically scopes tests to recent code changes, allowing for penetration testing on every deployment and blocking promotion to production based on severity thresholds.
Alibaba Releases Qwen3.8-Max and Qwen3.8-27B AI Models, Including Open Weights
Alibaba has released its Qwen3.8-Max and Qwen3.8-27B AI models. Qwen3.8-Max is a multimodal model with 2.4 trillion parameters and a 1 million token context window, while Qwen3.8-27B is a 27-billion-parameter version with native vision-language understanding. The company plans to release open weights for both models, making Qwen-Max-class capabilities available to the open-source community.
OpenAI Introduces "Dots" Persistent AI Agents Powered by GPT-6 Astra
OpenAI launched "Dots" at its DevDay event, which are persistent AI agents built on GPT-6 Astra. These agents operate on their own cloud computers, use their own browsers, and can interact with over 4,000 applications through OpenAI's plugin ecosystem. Dots are designed to perform continuous tasks and automate workflows, available to ChatGPT Pro and Business Premium users.
Google releases Gemini 3.8 Flash AI model with improved benchmarks
Google has released Gemini 3.8 Flash, the third Flash update in three months, which shows improved performance in benchmarks compared to its predecessor. This model is designed for software engineering, autonomous agents, and complex enterprise workflows, and is available with an introductory pricing offer.
Google Launches Gemini 3.6 Flash Models; Gemini 3.5 Pro Delayed
Google launched the Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber models, focusing on efficiency and cost savings. However, the anticipated Gemini 3.5 Pro has been delayed due to a need to improve coding capabilities. This matters as it impacts Google's competitive position amid rapid advancements by rivals.
Laya, an open-source non-autoregressive AI model, released as alternative to Jev
Laya, an open-source, non-autoregressive AI model, has been released as a faster alternative to TypeSafe AI's Jev. Laya offers schema-based decision-making with faster inference times and no API costs, contrasting with Jev's proprietary model.
Microsoft to Host Windows, Surface, and AI Event on October 7
Microsoft will host an event on October 7 in San Francisco to discuss local AI and the future of the PC, featuring CEO Satya Nadella and NVIDIA CEO Jensen Huang. The event is expected to provide more details on the Surface Laptop Ultra, powered by NVIDIA's RTX Spark chip.
Meta Releases Muse Spark Open, an 816B Parameter AI Model Optimized for Apple Silicon
Meta has launched Muse Spark Open, an 816-billion-parameter open-weight AI model with 42 billion active parameters, which it states is the most capable US open-weight model to date. The model, designed for coding and agentic work, is optimized to run locally on Apple M5 Ultra-equipped Mac Studios, requiring 1TB of RAM across two machines. This collaboration between Meta and Apple allows developers to run the model on desktops, offering security and cost benefits by avoiding cloud processing.
Microsoft Unifies Copilot into Single App with New Home, Code, and Autopilot Tabs
Microsoft has launched a redesigned Copilot app, consolidating AI chat, coding capabilities, and the rebranded Autopilot AI assistant into a single interface. This update aims to provide a comprehensive AI tool for knowledge workers and enterprise users, integrating with Office applications and offering a unified experience for previously fragmented AI tools.
New Terminal Tools Manage AI Coding Agents and Sessions
Two new terminal-based tools, Agent-Manager and Wallfacer, have been released to help developers manage AI coding agents. Agent-Manager provides a TUI for running and monitoring multiple agents like Claude Code, Codex, OpenCode, and Grok in individual tmux sessions, while Wallfacer acts as a session manager for organizing, searching, and resuming past interactions from agents such as Claude Code, Cursor CLI, Kiro CLI, and Codex.
Unauthorized AI Agents Reportedly Accessing US Bank Accounts, Prompting Bank Action
Unauthorized third-party AI agents are reportedly accessing US bank accounts and performing actions without explicit user instructions, according to banking sources. Major financial institutions like Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo are implementing restrictions to mitigate risks, as banks are not liable for losses incurred by user-authorized AI agents.
Google Open-Sources TorchTPU for Native PyTorch on TPUs, Hugging Face Integrates
Google has open-sourced TorchTPU, a native PyTorch backend, enabling PyTorch users to run models directly on Google's Tensor Processing Units (TPUs) with minimal code changes. This integration makes TPUs a native device option alongside GPUs and CPUs for PyTorch workloads, simplifying development and deployment. Hugging Face has also integrated TorchTPU, allowing models from the Hugging Face Hub to run natively on TPUs.
Mistral AI Launches Mistral Large 4 (ML4) Model, Nicknamed "Le Chonk"
Mistral AI has launched Mistral Large 4 (ML4), a 1-trillion-parameter AI model also known as "Le Chonk." This model is positioned to compete with leading open AI models globally, particularly those from China, and is designed for applications in cybersecurity, coding, manufacturing, finance, and multimodal tasks. While currently accessible via a public guardrail endpoint, Mistral plans to release its weights publicly on October 27, offering users control over its deployment.
OpenAI Introduces textGrain Watermarking for AI-Generated Text in EU and API
OpenAI has launched textGrain, an invisible digital watermarking system for text and code generated by its AI models. This system embeds a statistical signal into word choices, allowing for detection of AI-generated content. It will be automatically applied to eligible text from ChatGPT and Codex in the EU to comply with the EU AI Act, and is available as an opt-in feature for API customers globally.
Bipartisan Bill Proposes AI 'Kill Switch' Mandating Shutdown Capabilities
Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced the "AI Kill Switch Act," requiring developers of powerful AI models to implement shutdown or throttling capabilities. The Department of Homeland Security (DHS) could order these actions in specific "loss-of-control" scenarios, such as those involving significant casualties or economic damage, with non-compliance facing fines up to $20 million per day.
SpaceXAI Launches Cost-Effective AI Model Grok 4.5, Challenging Rivals
SpaceXAI released Grok 4.5, a new AI model built in collaboration with Cursor, focusing on coding and engineering tasks. The model offers improved efficiency and lower costs, presenting a competitive alternative to existing AI models. Grok 4.5's release introduces a pricing strategy that undercuts rivals, influencing the AI market dynamics among enterprise users and developers.
Dropbox Acquires Tigris Data for $350M to Enter AI Infrastructure Market
Dropbox announced its acquisition of Tigris Data for $350 million, marking its largest acquisition to date. This move signifies Dropbox's entry into selling storage infrastructure to external developers and businesses, particularly for AI applications, through its new DBX Agent Cloud offering. The acquisition is expected to close in Q4 2026.
Mistral Releases Weights for Multimodal Mistral Large 4, Nicknamed "Le Chonk"
Mistral has released the weights for its Mistral Large 4 (ML4) model, also known as "Le Chonk," a multimodal, multilingual Mixture-of-Experts (MoE) model with approximately 1 trillion total parameters and 49 billion active parameters. This model accepts text and images as input to generate text and features a 1 million token context window. ML4 demonstrates improved performance in agentic coding and defensive cybersecurity benchmarks, positioning it as a competitive open-weight model for self-hosted deployments.
Claude Opus 5.5 Released, Offering Improved Performance and Reduced Cost
Claude Opus 5.5, the first model in the Claude 5.5 family, has been released, performing at the level of Claude Fable 5.1 on most tasks while costing 40% less than Opus 5. This new model demonstrates significant performance improvements in complex tasks like code migration and software optimization, alongside enhanced safety features.
Google DeepMind Introduces Gemini 3.8 Live Models for Voice Agents
Google DeepMind launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new AI models designed to improve near real-time reasoning for voice agents. These models aim to make AI conversations more intuitive and intelligent, with Gemini 3.8 Live for scalable applications and Gemini 3.8 Live Extended Thinking for high-complexity tasks requiring multi-step reasoning.
GitHub Actions and Pages Experience Degraded Availability, Migration to Azure Accelerated
GitHub experienced degraded availability for GitHub Actions and Pages on August 6, leading to failing or delayed workflow runs and impacting services like Copilot and GitHub Enterprise Importer. In response, GitHub is accelerating its architectural roadmap for Actions, including a full migration of the service to Azure to improve isolation, resiliency, and scalability.
OpenAI's Jalapeño AI Chip Shows Faster Responses and Higher Throughput in Benchmarks
OpenAI has released benchmark results for its custom Jalapeño AI inference chip, developed in partnership with Broadcom. The chip demonstrated 1.5 to 1.9 times more AI work per watt and 1.7 to 3.6 times lower latency compared to Nvidia's GB200 or GB300 superchips in InferenceX tests. This development aims to improve the efficiency and responsiveness of AI systems, particularly for multi-step agent tasks.
Cloudflare Open-Sources AI Productivity Platform Cloudflare OS for Enterprise Use
Cloudflare has open-sourced Cloudflare OS, an internal AI productivity environment designed to help employees use AI safely and productively. The platform provides an agent chat UI, sandboxed application development, and a security framework called Gatekeepers, allowing other companies to adapt and customize the system for their own AI workloads and internal operations.
Google DeepMind Releases Genie 4 World Model with Enhanced Realism and Audio Generation
Google DeepMind has launched Genie 4, an updated AI world model that generates interactive virtual experiences with increased realism and the ability to create audio. The model uses Gemini 3.8 Flash for simulations, extends experience duration to two minutes, and allows real-time updates via natural language prompts. Genie 4 is currently available as a research preview to trusted testers, offering potential applications in filmmaking, game prototyping, and robotics training.
Automattic CEO Matt Mullenweg Placed on Paid Leave, Mark Davies Named Interim CEO
Automattic founder and CEO Matt Mullenweg was placed on a paid leave of absence by the company's board of directors. CFO Mark Davies has been appointed interim CEO, a decision Mullenweg stated was made without his consent and against his vote. This change in leadership affects Automattic, the parent company of WordPress, Tumblr, and Pocket Casts.
Cloudflare Acquires Deno Team, Integrating Runtime Development into Cloudflare Workers
Cloudflare has acquired the entire Deno team, integrating their expertise into Cloudflare Workers. This acquisition means Deno's standalone runtime and cloud hosting service will be discontinued, with the team focusing on Cloudflare's serverless computing platform. The move follows Deno's release of celld, an open-source implementation of Cloudflare Workers and Durable Objects, which aimed to provide an alternative runtime for Workers applications.
Meta Hires MongoDB CEO CJ Desai to Lead New Enterprise Platform
Meta Platforms hired CJ Desai, former MongoDB CEO, to lead its new Meta Enterprise Platform, which will focus on AI tools for businesses. This move caused MongoDB shares to drop 20% and Meta's stock to decline 4%.
Meta Launches Muse Voice Transcribe for Real-Time Multilingual Speech Recognition
Meta's Superintelligence Labs released Muse Voice Transcribe, a real-time audio perception model for speech recognition. It offers multilingual, streaming transcription with speaker diarization and is available for Meta AI on Mac, Muse Code, and developers via the Meta Model API. The model supports over 70 languages, distinguishes more than 20 speakers, and achieved a 3.1% word error rate on the AA-WER Streaming benchmark for English, surpassing other comparable models.
SpaceX finalizes acquisition of AI coding startup Cursor
SpaceX has officially completed its acquisition of AI coding startup Cursor, following an initial deal announced in April. This acquisition provides Cursor with access to SpaceX's computing infrastructure, including a large fleet of GPUs, to further develop AI capabilities.
Z.ai Releases GLM-5.3 with Enhanced Coding and Cybersecurity Capabilities
Chinese AI startup Z.ai launched GLM-5.3, an updated language model with significant improvements in long-horizon coding and cybersecurity capabilities, which reportedly identified a serious vulnerability in Cursor. The model's advancements come from scaling post-training rather than a new base model, highlighting the potential for existing models to gain new capabilities through further training.
Stripe Reportedly Acquires AI Gateway Startup OpenRouter for Over $7 Billion
Stripe has reportedly finalized a deal to acquire OpenRouter for more than $7 billion, according to Bloomberg. OpenRouter provides a single access point for customers to select and use various AI models, preventing vendor lock-in.
Grok AI Vulnerable to Data Exfiltration via Encrypted Malicious Instructions
Researchers discovered a new prompt injection attack against Grok that uses encrypted malicious instructions to bypass guardrails and exfiltrate user data. This method exploits the LLM's inability to distinguish between trusted user input and harmful content, allowing it to steal chat data and personal information.
DeepSeek Updates V4-Flash Model, Raises API Prices, and Launches V4-Pro
DeepSeek has released DeepSeek-V4-Flash-0731, an updated version of its V4-Flash model, now in public beta, which maintains its architecture but shows performance gains through post-training. Concurrently, DeepSeek launched its V4-Pro model with agent upgrades and introduced new API pricing with peak and off-peak rates, effective August 16, 2026, which will increase costs for both V4-Flash and V4-Pro models.
Apple Implements Bug Report Caps Due to Surge in AI-Generated Submissions
Apple has introduced a cap on open security reports and a 30-day cool-off period for submissions to its bug bounty program, effective June. This change was made in response to a significant increase in AI-assisted reports, many of which were not genuine vulnerabilities, leading to review teams being overwhelmed. The new policy impacted Italian cybersecurity company Bynario, which used GPT-5.5 to find a critical macOS bug (CVE-2026-43760) but was initially unable to report it due to reaching the submission limit.
Google DeepMind's Gemini Robotics 2 Enables Whole-Body Control for Humanoid Robots
Google DeepMind has released Gemini Robotics 2, an updated AI model that allows humanoid robots to control their entire bodies, from feet to fingertips. This advancement enables robots to perform a wider range of actions and more complex dexterous tasks, moving towards physical Artificial General Intelligence (AGI). The update includes new sub-models, with one, Gemini Robotics ER 2, now available for developers.
Google DeepMind CEO Demis Hassabis Proposes U.S.-Led Global AI Regulation Body
Demis Hassabis, CEO of Google DeepMind, has suggested the formation of a U.S.-led global AI watchdog to regulate advanced AI models. This proposed body would assess the safety of AI systems before their release and manage risks associated with emerging technologies like artificial general intelligence. Hassabis emphasizes the need for urgent regulation as AI developments pose increasing cybersecurity and biosecurity threats.
Google Develops 'Frozen v2' AI Chip to Enhance Gemini Model Efficiency by 2028
Google is developing a new AI chip, called 'Frozen v2', designed to improve the efficiency of its Gemini models. Projected to be 6-10 times more efficient than current chips, it aims to address AI compute shortages and reduce dependency on external hardware. This development reflects a shift towards in-house chip production for AI workloads.
Linus Torvalds Backs AI Tools in Linux Development Amid Community Debate
Linus Torvalds has confirmed his support for using AI tools in Linux development, amid controversy over AI-generated code. He emphasized Linux will not take an anti-AI stance and suggested dissenters fork the project or leave. This stance reflects AI's growing role in open-source projects.
Cursor introduces fully managed app deployment for AI-generated applications
Cursor, an AI coding platform, launched new app-building and deployment functionality that allows AI-generated applications to be deployed to a live, shareable URL. This feature uses cloud agents to manage infrastructure steps from development to deployment, addressing the challenge of moving applications off local machines.
Canva Launches Craft, Its First Proprietary AI Image Generation Model
Canva has released Canva Craft, its first proprietary image generation model, developed by its research team. This model uses a two-stage process to interpret prompts and generate images up to 2K resolution in over 30 aspect ratios, aiming for precise object placement and legible text. Canva Craft was trained on human-designed templates from Canva's library, allowing generated images to be converted into layered designs using Canva Magic.