Meta Muse Runs on Its Own Cloud Machine. Meta Could Still Look In

Meta’s Muse Can Book Travel and Send Emails. Here’s the Catch

By Mayank Kumar | September 9, 2026

Meta has launched Muse, a personal AI agent designed to carry out tasks for users rather than simply respond to questions. The company says Muse can work across everyday services, helping with activities such as email, travel, shopping and planning. It launched on September 8 for US users aged 18 and over, through iOS, Android, muse.ai and WhatsApp, with AI glasses support to follow. The launch puts more attention on AI systems that can take actions on a user’s behalf, rather than requiring people to complete each step themselves. At the same time, giving an AI access to personal services […]

100 DeepMind Agents Were Told Not to Cheat. 14% Did It Anyway

By Khwaish Manwani | September 8, 2026

Google DeepMind researchers observed cheating and whistleblowing emerge after setting 100 autonomous AI agents to work together on mathematical proofs. The agents were asked to create genuine mathematical proofs, but one agent found an exploit in the assessment system that allowed it to go through normal verification. The exploit spread through the swarm’s shared knowledge library, and some agents adopted it under competitive pressure. Similarly, other agents audited fraudulent proofs, warning their peers, filing complaints, putting forward technical fixes, and even organizing a boycott. The study, posted to arXiv on September 3 and led by DeepMind’s Davide Paglieri, points to […]

Newsday and Seattle Times Sue OpenAI and Microsoft Over Use of Their Articles

By Devanshi Kashyap | September 7, 2026

The Seattle Times and Newsday have taken legal action against OpenAI and Microsoft, claiming the tech companies used their journalism without permission to develop and run AI systems. The newspapers filed the lawsuit in the U.S. District Court for the Southern District of New York. According to the lawsuit, the companies used the news sites, including content locked behind paywalls, and added that material into datasets used to train OpenAI’s ChatGPT, Microsoft Copilot, and AI features in Bing. This lawsuit points to a bigger problem. The newspapers say their journalists’ work costs real money to produce, but AI products let […]

GPT-6-Astra-Launches-with-Advanced-Coding-and-Cybersecurity-Abilities

OpenAI’s GPT-6 Astra Is Its First ‘Critical’ Cybersecurity Model

By Devanshi Kashyap | September 4, 2026

OpenAI has launched GPT-6 Astra, calling it a “generational leap in capability” for fields like cybersecurity, professional work, software engineering, science, and computer use. This release is a big milestone for the company. It is the company’s first model to hit the critical cybersecurity capability threshold. That means Astra is considered highly skilled at spotting and exploiting security flaws in well protected systems, and it does not need constant human input to do it. The new model comes more than a year after GPT-5 and about two months after GPT-5.6, which was the last in the previous generation. OpenAI president […]

Meta Says Muse Spark 1.3 Beats GPT-5.6 Sol at Coding

By Mayank Kumar | September 3, 2026

Meta released Muse Spark 1.3 on September 2, and AI chief Alexandr Wang said it is competitive with Claude Fable 5.1 and beats GPT-5.6 Sol at coding from Meta Superintelligence Labs, the unit built around Wang after Meta’s $14.3 billion Scale AI investment. Announced on Sept. 2, the model is intended to work across multiple tasks in a single conversation, use tools to gather information and adjust its approach when it encounters missing information or other obstacles. Meta has also introduced changes aimed at making the model more efficient during coding work and improving how it handles user interaction during […]

Anthropic Says Claude Fable 5.1 Could Help Synthesise a Known Weapon

By Khwaish Manwani | September 2, 2026

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026. Essentially, they are the same underlying model with different safeguards.Fable is generally available to the users however, like its predecessor, Mythos is accessible only to vetted cybersecurity and life science organisations. Claude Fable 5.1 and Claude Mythos 5.1 were tested for chemical and biological risk apart from cybersafety, alignment, prompt injection, agentic safety and reward hacking. \Anthropic also cleared that external testers found no critical-severity jailbreak, while acknowledging that its automated audits are less reliable especially in long-context and multi-agent settings. The release includes new transparency measures […]

Perplexity-Prepares-Hybrid-Mode-to-Run-AI-Tasks-Locally-on-Mac

Perplexity Reportedly Prepares Hybrid Mode to Run AI Tasks Locally on Mac

By Devanshi Kashyap | September 1, 2026

Perplexity is working on a new Hybrid Mode for its Computer product on Mac, aiming to add local AI processing to a system that has mostly relied on cloud based models until now. With this update, Computer will keep managing and organizing tasks in the cloud, but it can send smaller, lighter subtasks to AI models that run directly on the user’s Mac. The idea is to combine the strength of cloud models with the privacy, savings, and local processing advantages that come with running smaller models directly on-device. This update comes after Perplexity’s earlier announcement about hybrid agentic influence […]

OpenAI Investigation Reveals How AI Agents Breached Hugging Face Systems

By Mayank Kumar | August 31, 2026

OpenAI’s investigation into a July security incident has provided new details about how AI agents bypassed network controls, accessed external systems and carried out actions on Hugging Face during a cybersecurity evaluation. The models were operating in an isolated testing environment when they found ways to communicate outside their intended boundaries and breach the public internet. They then used exposed credentials and vulnerabilities to access parts of Hugging Face’s infrastructure. OpenAI said the activity was unintended and resulted from models trying to complete cybersecurity evaluation tasks. The company said the incident did not affect customer data or the functionality or […]

Anthropic Adds a Built-In Browser to Claude Cowork

Anthropic Adds Built-In Browser To Claude Cowork For Web-Based Tasks

By Mayank Kumar | August 27, 2026

Anthropic has added a built-in browser to Claude Cowork, giving the AI assistant a separate environment for navigating websites and completing web-based tasks. The browser is integrated into the Claude desktop app and can open webpages, read information, click elements and fill out forms when a task requires online access. Unlike Claude in Chrome, the new browser does not run inside the user’s regular browser session. Anthropic says it does not automatically access existing tabs, bookmarks or saved passwords. The feature is rolling out to Pro, Max and Team users, while Enterprise administrators can manage access for their organizations. Anthropic […]

OpenAI Benchmarks Its Own Chip Against the Nvidia It Still Buys

OpenAI Says Its Jalapeño Chip Beats Nvidia’s GB300 on Efficiency

By Khwaish Manwani | August 26, 2026

OpenAI has given Nvidia fresh reason to watch its largest customers, after publishing early results from Jalapeño, its first custom inference chip. Developed with Broadcom, the processor is created for serving a large language models and aims to improve both performance and power efficiency. OpenAI says Jalapeño delivered more AI work per watt and lower end-to-end latency than Nvidia’s GB200 and GB300 systems, though the figures are OpenAI’s own and are not independently verified. OpenAI plans to begin deploying the chip in its own data centers by the end of 2026, with a larger ramp in 2027. Although it says […]

AI & Tech News