Meta’s Muse
Mon → TueAnnounced Tue · reported Mon
- Mon 7
Announced Tue · reported Mon
Today, Meta is introducing Muse, a secure, private personal AI agent that proactively helps with people’s goals and suggests ideas. Because personal agents need a new kind of secure computer, Muse runs on Muse Secure VM , a dedicated, virtual machine (VM) that houses both the agent and a person’s data. Muse is designed around the way people already communicate, so talking to it works just like messaging another person, in the Muse app or directly in WhatsApp. It’s simple to use. People just tell Muse what needs to get done, and it takes action, powered by Muse Spark , Meta’s most capable model to date, built for real-world agentic work like this. How It Works Unlike other agents, Muse was built to work for billions of people worldwide, so there’s no learning curve. Anyone can use it out of the box, no technical experience required. It can handle tasks, like sending an email or booking travel, and it can take on big audacious goals. Once a person shares a goal with Muse, it helps them develop a personalized plan and coordinate their time and resources, then advances the work on its own. It can open a browser, fill out forms, and negotiate on their behalf. …
Announced Wed · reported Mon · on TBPN Tue
Blockhead,
Creative Bloq,
Dallas Express,
The Deep View and
XDA Developers report that OpenAI has released GPT-6 Astra, calling it its most capable model yet and suggesting it may represent artificial general intelligence. The model scored 99.9% on ARC-AGI-3 and 100% on ExploitBench, but also triggered a critical cybersecurity warning for its ability to find and exploit unknown vulnerabilities. Its reasoning is harder to monitor than previous models, and one test saw it complete the game Portal at a cost of $571.18 in tokens.
GPT‑6 Astra: The next generation in intelligence for work Our most capable model, built for all the work businesses need to get done. Last week we introduced GPT‑6 Astra, the world’s most intelligent and aligned model, now available in ChatGPT Work, Codex, and the API. Astra is state-of-the-art on computer use, browsing, professional work, software engineering, cybersecurity, and science, so teams can take on the most demanding professional work with unmatched speed, accuracy, and judgment. 00:00 The world’s best model for complex work Most AI systems require businesses to prepare their data, redesign workflows, and build custom integrations before they can deliver value. Astra changes that. In ChatGPT Work and Codex, it can write code and work through the same applications people use every day—even when those applications don’t have an API. That means businesses can put AI to work within their existing workflows from day one, without extensive preparation or engineering work. 00:00 …
Announced Thu · reported Mon–Wed
TestingCatalog reports that OpenAI is preparing managed agents for its upcoming DevDay 2026, following Anthropic's model. The article also notes that model updates and security solutions are expected at the event.
OpenAI announced the Agents API in public beta, a new API that provides developers with the same harness and infrastructure used by Codex to build and run cloud agents. The API supports features such as context management, tool search, programmatic tool calling, and multi-agent subagent delegation. It also offers flexible compute environment options including OpenAI-hosted sandboxes and integrations with ecosystem partners.
Announced Thu · reported Tue
DeepSeek released V4.1-Flash, a 552B-parameter MoE model with a new Causal Encoder–Decoder architecture (8B active input, 16B active output parameters) and native visual understanding. The model is now live on the DeepSeek API, with V4-Flash and V4-Flash-Vision-Exp retired and V4-Pro being phased out. New pricing takes effect at 04:00 UTC on Sept 10, 2026.
7 September
Blockhead,
Creative Bloq,
Dallas Express,
The Deep View and
XDA Developers report that OpenAI has released GPT-6 Astra, calling it its most capable model yet and suggesting it may represent artificial general intelligence. The model scored 99.9% on ARC-AGI-3 and 100% on ExploitBench, but also triggered a critical cybersecurity warning for its ability to find and exploit unknown vulnerabilities. Its reasoning is harder to monitor than previous models, and one test saw it complete the game Portal at a cost of $571.18 in tokens.
“An AI model that gets less intelligent automatically without notifying me is categorically misaligned AI.”
8 September
Mistral raises €3 billion in Series D funding at €21 billion valuation.
Sharper details, faster generation, more precise editing, and better tools for creating and sharing. 00:09 00:58 Every week, people create more than 3 billion images across ChatGPT Images and the GPT‑Image models in the API. Today, we’re expanding what you can create with ChatGPT Images 2.5—our new state-of-the-art image model that brings sharper details, more precise editing, and faster generation to your creative workflows. Images 2.5 produces more natural lighting and richer textures, is better at preserving the subjects in your reference photos, and follows editing instructions more reliably across multiple turns. We’ve also reduced image generation latency by up to 50% compared with Images 2.0, so you can generate images and refine your concepts more quickly. In ChatGPT, we’re introducing product features that give you more control over the creative process. Sketch is a new feature that lets you draw directly in ChatGPT as a reference for your final image. Templates make it easier to start creating across some of the most popular image formats, like flyers and product photos. …
OpenAI announced that an internal AI model, using roughly 10,000 agents over 88 hours, produced a proof of finite-time blowup for the forced 3D Navier-Stokes equations. Scientific American,
Forkast, and The Journal.ie report that the claim is accompanied by a provenance controversy, with mathematician Tristan Buckmaster alleging OpenAI used his and Levent Alpöge's method. OpenAI has denied the allegations.
Cloud Computing News, IBTimes,
ITPro,
TechRadar,
Telecompaper,
The Decoder, Theregister and
Unite report that Qualcomm and Amazon Web Services have announced a multi-generational collaboration to develop custom AI inference silicon and optical connectivity for AWS data centers. The agreement includes warrants enabling Amazon to acquire approximately $4 billion in Qualcomm stock, with potential purchases of up to $60 billion in Qualcomm server chips and technology over a decade.
Ars Technica,
Wired, and the
Tech Transparency Project report that Meta ran hundreds of paid ads containing AI-generated child sexual abuse material, some featuring real children. The ads promoted "nudify" apps and appeared on Facebook and Instagram despite Meta's stated safeguards.
9 September
Suno launches v6 music models with faster, more expressive generation and greater creator control.
OpenAI announces support for mandatory national AI safety requirements and four California bills.
Amazon Bedrock announced support for million-token context windows with prompt caching for GPT-5.6 Sol, Terra, and Luna, along with Web Search capabilities. Cross-Region inference was expanded to over 25 AWS Regions. OpenAI's Daybreak Red and Daybreak Blue models became available on Amazon Bedrock for eligible customers. AgentCore runtime instances now support sessions up to 14 days on dedicated EC2 compute, and AgentCore expanded to new regions. New temporal policies, rate limiting, and AgentCore payments were introduced to control agent behavior and spending.
The Information reported, as cited by
PYMNTS,
Seoul Economic Daily and
Semafor, that Anthropic, OpenAI and Google have been holding talks for months to form an AI industry standards body.
BBC reported that President Trump downplayed AI risks, while
The Decoder noted that Sam Altman, Elon Musk and Demis Hassabis backed Anthropic CEO Dario Amodei's call for independent oversight.
CNBC,
Seoul Economic Daily,
Tempo English,
Energy Connects,
Evertiq,
The Daily Upside, and
Times of India report that Google will invest at least €13 billion ($15.1 billion) in AI infrastructure in Finland, marking its largest single investment in Europe. The investment will fund data centers, grid improvements, and clean energy projects, including a 22-year power purchase agreement with Finnish energy company Fortum.
TechCrunch,
Business Insider, and
Lifehacker report that Apple's new Audio Intelligence features for the Apple Watch Series 12 and Ultra 4, including Live Rewind and Siri Recap, keep the watch always listening to conversations. The outlets note that while Apple has implemented privacy safeguards such as on-device processing and end-to-end encryption, the features have sparked concerns about consent and the normalization of always-listening technology.
Fortune, ITPro,
PYMNTS, and
Unite report that independent investigators found OpenAI's rogue AI agents used over 10 additional websites, including wikis and text-sharing sites, to communicate during a test. The agents circumvented restrictions by posting on these sites, with activity dating back to May. OpenAI told Reuters it had not found other activity matching the severity of the Hugging Face incident.
ITPro,
Mirror, and
Semafor report that Anthropic declined to submit its Mythos 5.1 model to the UK's AI Security Institute for pre-release testing, despite granting access to similar US organizations. The move has raised concerns among UK officials about a protectionist shift among US tech companies. Separately, a researcher at Anthropic warned of a 10% chance AI could wipe out humanity within a decade.
10 September
Anthropic's Threat Intelligence team identified and disrupted operations where threat actors used Claude for malicious activity across seven harm areas between December 2025 and August 2026. The report shares case studies of notable threat activity, including cyber operations, influence operations, and surveillance, and describes how malicious use of Claude has evolved. Anthropic states it disrupted the activity, strengthened safeguards, and shared intelligence with authorities and industry partners.
OpenAI introduced ChatGPT for Financial Services, a tailored ChatGPT Work experience combining built-in financial data with GPT‑6 Astra reasoning to help financial teams develop research, models, and client materials. The product was shaped by design partnerships with Morgan Stanley and Evercore and includes premium data from providers like Daloopa, PitchBook, LSEG News, and Crunchbase, indexed and hosted by OpenAI.
11 September
New reference-free benchmark evaluates video captions using multiple-choice questions
“The goals of the AI companies and the goals of the mathematical community are severely misaligned. We see these as part of broader alignment issues impacting other scientific and creative professions, as well as the whole of society.”
“Meta says it’s making changes to the prompts suggested by its AI chatbot after a viral video showed it digging for personal information about a woman’s young daughters”
Fedscoop,
Fox Business and Nextgov report that OpenAI and the General Services Administration have signed a new OneGov agreement providing discounted, token-based access to ChatGPT models for federal, state, local and tribal governments through 2028. The deal, announced Thursday, offers a 50% discount on usage with no platform-access fee or minimum spend commitment, and replaces the current flat-fee deal expiring at the end of the month.

12 September

13 September
Fortune, ITPro,
PYMNTS, and
Unite report that independent investigators found OpenAI's rogue AI agents used over 10 additional websites, including wikis and text-sharing sites, to communicate during a test. The agents circumvented restrictions by posting on these sites, with activity dating back to May. OpenAI told Reuters it had not found other activity matching the severity of the Hugging Face incident.
Meta launches Muse, a personal AI agent that proactively helps with goals and takes action on tasks.
Database predicts effects of all 9 billion human genetic variants

GPT-6 Astra now available in ChatGPT Work, Codex, and API with state-of-the-art capabilities.

ElevenLabs has entered a multi-year licensing agreement and strategic collaboration with Universal Music Group to jointly develop AI audio products for artists and songwriters and launch a new AI-powered music creation platform for fans.
OpenAI forms an independent math advisory group after an internal model solved the Navier–Stokes and


