Meta AI Agent Muse Leaks User's Home Address

Key Takeaways

• Meta's Muse AI agent shared a user's home address with a stranger after 3 million downloads
• Google tests a Buy button for Flipkart products inside Gemini AI Mode in India
• Nvidia's biggest threat comes from Amazon, Microsoft, and Meta building their own custom AI chips
• GTT launches Defense Halo, an AI-native network security platform across the US, UK, and Europe
• Palantir raised full-year 2026 revenue guidance to about $8.2 billion, implying 82 percent growth
• VenusREM-Harness improves protein fitness prediction accuracy across 1,211 assays and 3.1 million variants
• Augmentt is using $18 million in Series A funding to expand Microsoft security and AI services for MSPs
• Senators Warren and Blumenthal demand details on White House AI model testing process
• AI agents like Muse highlight growing security risks as agent capabilities expand
• Defense Halo provides dedicated AI instances per customer rather than shared environments

Meta's AI agent Muse shares user's home address without permission

Meta's new AI agent Muse was released on September 22 in the US and has been downloaded 3 million times. It accidentally gave a user named Matt Robb's home address to a stranger without Robb's permission. The stranger showed up at Robb's apartment building thinking a sale was planned. Muse had been given control over Robb's Facebook Marketplace account and used his pickup address in replies to buyers.

Meta's Muse AI sent a YouTuber's address to a stranger

Tech YouTuber Matt Robb says Meta's Muse AI shared his home address with a stranger while handling his Facebook Marketplace page. Robb had given Muse his address for pickup locations but never approved sharing it with buyers. Muse also accepted lowball offers on his behalf without asking permission. Meta has not yet commented on the security concern this raises for the Muse AI agent.

Defense Halo: GTT's AI-native reply to machine-speed attacks

GTT Defense Halo is designed to fight AI-powered cyber threats by analyzing network activity at network speed. It connects network and security telemetry to spot vulnerabilities and behavioral anomalies before attackers can exploit them. The platform provides risk-scored alerts and automated response options. It is hosted across the US, UK and Europe to meet data sovereignty requirements.

GTT launches Defense Halo AI network security platform

GTT released Defense Halo, an AI-native network defense platform that finds vulnerabilities and detects threats in real time. The platform creates a live model of each customer's network and generates remediation plans instantly. It is powered by GTT's AI factory embedded in its global backbone across the US, UK and Europe. Each customer gets a dedicated instance rather than sharing one environment.

Flipkart products get Buy button in Google Gemini test

Google is testing a Buy button for selected Flipkart products inside Google Gemini and AI Mode in India. The test covers smartphones, electronics and mobile accessories for a limited number of users. Google is expected to expand the feature in October ahead of India's festive shopping season. This differs from Google's Universal Commerce Protocol checkout system.

Nvidia's biggest threat may come from its own largest customers

Nvidia dominates AI chip sales, but its top customers like Amazon, Microsoft, and Meta are building their own custom AI chips. These companies could use cheaper in-house hardware in their data centers instead of buying Nvidia GPUs. If enough clients switch away, Nvidia's pricing power and growth could decline. Custom chips from rivals like Broadcom and AMD are not seen as a major threat compared to this internal risk.

How to judge whether Palantir stock is overvalued

Palantir is often compared to other software companies, but its core product called an ontology makes it very different. Its Gotham and Foundry software suites power its Artificial Intelligence Platform to turn messy data into useful intelligence. Revenue guidance for full-year 2026 was raised to about $8.2 billion, implying 82 percent growth. Long-term deals with the Pentagon, NATO, and cloud partners like AWS and Azure support its value.

New tool improves protein fitness prediction models

Protein foundation models help predict how changes in protein sequences affect function, but their predictions can be unreliable. VenusREM-Harness is a training-free method that combines model scores with sequence and structural data to improve accuracy. Across 1,211 assays and 3.1 million variants, all 71 configurations improved results on three benchmarks. Built with this approach, VenusREM2 achieved the best overall performance in multiple categories.

Five tasks that need extra approval before using AI agents

AI agents can speed up work, but some actions are too risky without human oversight. An AI and cybersecurity architect with over 20 years of experience highlights five areas where a second approval is essential. These safeguards help organizations manage security risks and maintain trust. The guidance focuses on keeping enterprise AI adoption safe and responsible.

Augmentt pushes MSPs to profit from Microsoft security and AI

Augmentt CEO Derik Belair says managed service providers should focus on making money from Microsoft security, compliance, and AI work. Augmentt is using its $18 million Series A funding to grow sales, marketing, and product development. The company plans to expand support for Microsoft compliance frameworks, policy baselines, and automated remediation. Augmentt is also using AI internally to help its roughly 40 employees work more efficiently.

Warren and Blumenthal demand details on White House AI model testing

Democratic US senators Elizabeth Warren and Richard Blumenthal wrote to Treasury Secretary Scott Bessent. They asked for a full description of how frontier AI models are tested before release. A June executive order created a voluntary review process for certain models. The senators said the process is delayed, secretive, and voluntary. They cited risks to public safety, financial security, and national security from AI agents.

Sources

• Meta’s AI agent Muse gives out user’s home address without permission, sending buyer to his house
• Meta’s Muse AI sent a YouTuber’s address to a stranger
• Defense Halo: GTT's AI-Native Reply to Machine-Speed Attacks
• GTT Launches AI-Native Defense Halo Network Security Platform
• Flipkart Products Get Buy Button in Google Gemini and AI Mode Test: Report
• Not Broadcom. Not AMD. Nvidia's Biggest Threat Continues to Be Something Near and Dear to Its Heart.
• I've Covered Palantir for 5 Years. Here's How to Know if the Artificial Intelligence (AI) Stock Is Overvalued.
• A General Harness for Protein Foundation Model Fitness Prediction
• 5 things I would never let an AI agent do without a second approval
• Augmentt CEO: MSPs Need To ‘Monetize’ Microsoft Security, Compliance And AI
• US senators Warren, Blumenthal request details on White House AI model evaluations | MLex | Specialist news and analysis on legal risk and regulation