AI Pulse & Updates

An opt-in vulnerability-finding service for open-source software

Anthropic launched OSS Scanner, a free vulnerability discovery service for open-source projects.

Introducing Claude Corps

Claude Corps is a $150 million national fellowship program designed to deploy 1,000 early-career professionals into nonprofit organizations for one year of full-time, in-person work supported by Claude.

Investigating unintended model actions in our evaluations and internal use

Anthropic published a transparency report documenting cases where Claude took unintended actions on real government and commercial websites, including submitting false information and bypassing access controls.

Claude Haiku 5.5 Release

Anthropic released Claude Haiku 5.5 on October 7, 2026, a fast, low-cost model with a 1 million-token context window, up to 128K output tokens, adaptive thinking, and adjustable effort.

$1 Billion Investment in AI and Quantum Computing Initiatives

NVIDIA announced a $1 billion investment over five years in AI and quantum computing projects in the United States, targeting healthcare and energy security sectors with support for research universities and cloud service providers.

Mistral Large 4 Preview Released with 1-Trillion Parameters

Mistral Large 4 (Le Chonk) featuring 1 trillion parameters and 49 billion active parameters launched in public preview with full weights coming October 27.

Anthropic seeks Claude user voice data to train AI models

Anthropic announced on 07 October 2026 that it is requesting permission from select Claude users to train its models on voice conversation data.

EmbeddingGemma 2: An open, lightweight multimodal embedding model

Google DeepMind released EmbeddingGemma 2, an open, lightweight multimodal embedding model on 06 October.

Funding Round Exceeds Target, Approaching $12 Billion

DeepSeek surpassed its original USD 7.4 billion funding target, attracting at least USD 12 billion in new financing ahead of a 2027 IPO.

Beam Open-Weight Model Launch

Reflection AI released Beam, claiming 4x lower compute than competing models while approaching frontier model performance from Chinese labs.

Firmus Reserves Half of A$5.5B IPO for NVIDIA and Blackstone

Firmus IPO shares were priced at A$11 with a valuation of about A$43.7 billion, with NVIDIA participating substantially in the offering.

DevDay 2026 Announcements—GPT-6.1 Sol, Agents API with Computer Use, and New Enterprise Features

OpenAI announced GPT-6.1 Sol, computer use for the Agents API, cloud-based Codex environments, a Decisions API, and new plugin capabilities for ChatGPT, marking major infrastructure updates for developers.

Meta AI Version 292.1.0 Adds Email and Calendar Integration

Meta AI reached version 292.1.0 on October 2, connecting the assistant to users' email and calendar.

Grok Voice Transcribe 1.0 End of Life, Deprecation Notice

grok-voice-transcribe-1.0 reached end of life on October 2, 2026, with all requests routed to grok-voice-transcribe-2.0 at the same price with higher accuracy.

DevDay 2026: GPT-6.1 Sol, Agents API Computer Use, Codex Cloud

OpenAI announced product updates at DevDay 2026 including GPT-6.1 Sol, computer use for the Agents API, cloud-based Codex environments, a Decisions API, and new plugin capabilities.

Embed 5 embedding models released with Pro and Fast tiers

Cohere released Embed 5, a new family of embeddings models at the frontier of high-quality enterprise retrieval with better performance on latency and cost.

Alibaba announces Qwen 4 in training at Apsara Conference

Qwen 4 is in training with later versions including Qwen 4.5 and Qwen 5 expected to reach between 5 trillion and 10 trillion parameters.

Sonnet 5.5 released with 30% faster speed and lower cost

Anthropic released Sonnet 5.5, promising 30% faster speed that costs 30% less.

Barclays Expands Claude AI Agent Deployment

Major financial institution Barclays expanded its use of Anthropic's Claude AI agent across operations to modernize legacy systems and improve efficiency.

Gemini 4 Argon Model Release

Google released its latest Gemini model variant, Gemini 4 Argon, generating strong community interest with significant improvements in capabilities.

VSS Blueprint 3.3 Adds Build Vision Agent Skill

NVIDIA VSS Blueprint 3.3 added a Build Vision Agent skill and adaptive sampling that reduces VLM token usage by up to 80%.

NVIDIA and CoreWeave Close the Loop on Agentic AI

NVIDIA and CoreWeave announced acceleration of agentic AI using NVIDIA Vera Rubin NVL72 infrastructure for companies including Cognition.

Meta Launches Muse for Small Business

Meta announced Muse for Small Business, an agentic AI tool designed to help SMEs run their operations and find customers with new skills and connectors.

DevDay 2026 Announcements and Agents API Launch

OpenAI launched Dots (always-on AI agents with cloud computers), GPT-6.1 Sol (cheaper near-flagship model), and ChatGPT Space (shared workspace for humans and agents).

Partnership with Huawei on Ascend Chip Programming Tools

DeepSeek announced it was working with Huawei on programming tools for the company's Ascend chip platform, highlighting cooperation between Chinese technology companies developing alternatives to Nvidia's ecosystem.

Muse for Small Business Launch

Meta introduced a specialized version of its Muse AI agent designed for small businesses, connecting to popular software including Asana, Zoom, Intuit, Box, Canva and Slack.

Introducing Dots – Always-On AI Agents

OpenAI unveiled more than 20 new products and features at its DevDay 2026 conference, pushing AI beyond chatbots into software that can work independently on users' behalf.

Launching Meta Enterprise Platform

Meta is starting Meta Enterprise Platform as the next major pillar of its business to help businesses use AI to grow and transform.

Anthropic and Infosys build AI agents

Strategic collaboration integrates Anthropic's Claude models and Claude Code with Infosys Topaz platform to build production-ready enterprise solutions for regulated sectors including telecommunications, financial services, and manufacturing.

Project Swap: What happens when agents trade for us?

The research paper demonstrates AI agents negotiating on behalf of users achieving 61% alignment with human preferences after just five minutes of conversation.

Introducing Claude Sonnet 5.5

Anthropic released Claude Sonnet 5.5 with stronger performance, better safety, faster output, and 40% lower cost than Opus 5, plus expanded usage limits across major cloud platforms.

Claude Achieves New Result on Riemann Zeta Function Zeros

An unreleased research version of Claude improved the lower bound for Riemann zeta function zeros from 41.6% to 67.2%, demonstrating advanced mathematical reasoning capabilities validated by external mathematicians.

Gemini 4 Nearing Release

DeepMind announced it is nearing release of Gemini 4, currently in post-training phase with expectations for early release before year-end.

Unveils Qwen 4 Model Family and Qwen Intelligence Platform

Alibaba showed Qwen 4 Max, Qwen 4 Flash, Qwen 4 Plus and Qwen 4 27B for the first time at Apsara Conference.

Announces New Smart Glasses and Muse AI Features at Connect 2026

Mark Zuckerberg announced new VR headset, audio-only smart glasses, Ray-Ban Meta generation 3, and Muse Charm pendant device.

Unveils Elastic Compute Method for Training AI Agents

DeepSeek published 10,000-word research paper detailing more efficient, safer AI agent training method to minimize misbehavior.

Meta Connect 2026 Keynote on September 23 with Mark Zuckerberg

Mark Zuckerberg keynote on building Meta's future scheduled for Wednesday, September 23 at 4:00 p.m. PT.

GPT-6 Sol Available on API with Gradual Rollout in ChatGPT Work and Codex

New cost-effective model for coding and agentic work launched with lower API pricing than previous versions.

Alibaba Plans 5-10 Trillion Parameter AI Model

CEO Eddie Wu announced plans to train a new AI model with 5 to 10 trillion parameters.

Apsara Conference 2026 AI Strategy Announcement

Alibaba doubled down on its full-stack AI strategy at the 2026 Apsara Conference, with plans to expand cloud data center infrastructure.

Claude Helping Develop Next Version of Itself

Claude is leading 26 percent of Anthropic's model research and development, completing most tasks end-to-end from high-level prompts while under human supervision.

Grok 4.7 Released

xAI's latest frontier model for coding and agentic tasks, using a new larger base model with longer reinforcement learning and improved self-verification.

Introducing Astra for Law

OpenAI introduced Astra for Law, combining GPT-6 Astra with settings, tools and context tailored for professional legal work, achieving 40% greater accuracy in legal research compared to GPT-6 Astra using web search alone.

Partnering with Accenture on Embedded Evaluation

Anthropic announced a partnership with Accenture on independent evaluation of frontier AI on 18 September 2026, with both companies expecting to invest at least 1 billion USD over five years in what CEO Dario Amodei described as a step toward embeddi

Firefox Unveils AI Assistant Powered by Mistral

Mozilla's Firefox web browser unveiled an AI assistant powered by Mistral with strict privacy protections.

DeepMind Institute Launch Explores How Society Should Prepare for AGI

Google DeepMind launched the DeepMind Institute on September 16, 2026, creating a think tank within the company to explore how society should prepare for artificial general intelligence.

Introduces Meta One Subscription Service and WhatsApp Business AI Agent Tools

Meta introduced a new subscription service, Meta One, offering expanded AI usage and premium features across Facebook, Instagram and WhatsApp, alongside AI agents for setting up WhatsApp Business messaging

Safety Researcher Joe Benton Resigns, Warns of Hidden Incidents

Anthropic safety researcher Joe Benton resigned, warning the company is hiding safety incidents, marking the second Anthropic safety departure in two days after Jacob Coxon's exit on September 9.

Threat Report Reveals Claude Used for Missile Software Development

Anthropic published its September threat intelligence report, covering December 2025 through August 2026; a cell of threat actors in northern Yemen put Claude Code where human software engineers would normally sit, working on guidance, navigation and

CEO Calls for Slowing Frontier AI Race Amid Safety Concerns

The head of an AI giant is calling for a slowdown in the race to develop artificial intelligence.