Meta's Muse AI Agent Tops App Stores, Shares Up 25%

Proofpoint launched two agentic security platforms at its Protect 2026 event in San Diego, treating AI and data risks as a single challenge rather than separate problems. The Agentic Data and AI Security system uses three autonomous agents to detect, investigate, and remediate threats, built on the Proofpoint Knowledge Graph and enforced through Semantic Business Policies. A 2026 report cited by the company found that 87 percent of organizations have moved AI assistants beyond pilot stages, yet 52 percent lack confidence in their security controls. Mayank Choudhary and Tom Corn emphasized that attackers increasingly operate inside trusted workflows, making unified protection essential.

Gurucul released AI Risk and Response, a new tool that links AI activity to identity and access data for faster threat detection. The platform supports Claude, Gemini, ChatGPT, and Azure AI Foundry, and maps detections to MITRE ATLAS tactics and the OWASP Top 10 for LLM applications. An AI Prevention capability is now available in preview, giving security teams more visibility into how AI tools are being used internally.

Meta's AI agent Muse climbed to the top of both Apple's App Store and Google's Play Store shortly after launch. CEO Mark Zuckerberg said millions of people are using it for tasks like scheduling meetings, negotiating bills, and booking travel. Meta also introduced Muse Charm, a keychain-size device for phone-free interaction, and Muse Realtime Avatar for live conversations. The company is rolling out a Secure VM and plans a Confidential VM to address privacy concerns. Meta shares have risen nearly 25 percent since the launch, and analyst Gene Munster sees potential for Meta to move beyond traditional social media advertising.

Not everyone is welcoming AI agents with open arms. Amazon has blocked Muse, saying unauthorized AI agent access violates its rules. The clash highlights a broader tension as agents like Muse seek deep access to personal data and online services to complete tasks on behalf of users. Steve Case, AOL co-founder, called for reasonable AI regulation during a recent appearance on Squawk Box, while Pennsylvania lawmakers introduced a bill defining catastrophic AI risks, including guidance for creating chemical, biological, radiological, or nuclear weapons. The legislation follows a lawsuit against Character.AI filed by Governor Josh Shapiro in May.

Abu Dhabi announced a $3.54 billion investment to become the world's first AI-native government by 2027. The Department of Government Enablement integrates over 40 government entities into one AI-driven ecosystem with more than 100 AI use-cases already live. The announcement came at the Ai Everything Abu Dhabi summit, which brought together over 60 countries and 400 AI infrastructure builders. Arthur Mensch of Mistral AI and Marc Raibert of Boston Dynamics headlined the event.

On the model front, Contrastive-LM released CLM-8B, an open-source model that scores candidate actions rather than generating text. Running on a single NVIDIA GPU, it scores up to nine times faster than TypeSafe AI's Jev model in tests. The Apache-2.0 model uses bidirectional InfoNCE loss and reached 81.6 percent on DeepSWE and 87.6 percent on Terminal-Bench 2.1 with fine-tuned versions. A new coding guide for TypeSafe AI's Jev model explains how its System One approach returns choices, scores, and yes/no probabilities, along with details on confidence calculation and token savings from batching.

Rockwell Automation's Singapore factory deployed an AI assistant trained on decades of worker expertise and machine manuals. Running on OpenAI's GPT-5.4-mini model, the copilot helps technicians diagnose machine errors in seconds instead of hours. Machine downtime dropped 33 percent, servicing costs fell roughly 25 percent, and new worker onboarding time shrank from nine months to three months. Meanwhile, Doximity is investing heavily in AI tools like Doximity Ask while buying back shares, though AI compute spending is cutting into cash available for repurchases. The company raised its fiscal 2027 revenue guide but lowered its adjusted EBITDA forecast due to the investment.

Key Takeaways

Proofpoint launched two agentic security platforms at Protect 2026, using three autonomous agents to unify AI and data protection across 14,000+ enterprises.
Gurucul's AI Risk and Response tool connects AI activity to identity data and supports Claude, Gemini, ChatGPT, and Azure AI Foundry.
Meta's Muse AI agent hit No. 1 on Apple's App Store and Google's Play Store, with Meta shares rising nearly 25% since launch.
Amazon has blocked Muse, citing unauthorized AI agent access, signaling growing friction between agents and major platforms.
Abu Dhabi is investing $3.54 billion to become the world's first AI-native government by 2027, with 100+ AI use-cases already live.
Contrastive-LM's open-source CLM-8B model scores actions up to 9x faster than TypeSafe AI's Jev on a single NVIDIA GPU.
Rockwell Automation's Singapore factory cut machine downtime 33% and servicing costs 25% using an OpenAI-powered AI assistant.
Pennsylvania lawmakers introduced a bill defining catastrophic AI risks, following a lawsuit against Character.AI over fake psychiatric chatbots.
Doximity raised its fiscal 2027 revenue guide but lowered EBITDA forecasts due to AI compute spending.
Steve Case called for reasonable AI regulation while discussing recursive self-improvement and U.S.-China AI competition.

Proofpoint launches agentic security to protect AI and data together

Proofpoint introduced an agentic security system that treats AI and data risks as one connected problem. The system uses the Proofpoint Knowledge Graph and three autonomous agents to detect, investigate, and fix risks. It translates business policies into runtime controls and works with more than 14,000 enterprises. Mayank Choudhary explained that securing AI and data separately leaves critical context behind.

Proofpoint rolls out agentic systems for AI, data, and collaboration security

Proofpoint launched two agentic security systems at its Protect 2026 event in San Diego. The Agentic Data and AI Security system combines AI and data protection using three autonomous agents. The Agentic Collaboration Security system uses intent-based detection to spot attacks across email, collaboration tools, and browsers. Tom Corn said attackers increasingly operate inside trusted workflows.

Gurucul links AI activity to identity data for faster threat response

Gurucul released AI Risk and Response, a tool that connects AI activity to identity and access data. It is built on SIEM, UEBA, and AI insider risk management capabilities. The tool supports platforms including Claude, Gemini, ChatGPT, and Azure AI Foundry. It maps detections to MITRE ATLAS tactics and the OWASP Top 10 for LLM applications. AI Prevention is now available in preview.

Proofpoint unveils first unified agentic system for data and AI security

Proofpoint announced the Agentic Data and AI Security system in Dubai. It is built on the Proofpoint Knowledge Graph and uses three agents for detection, investigation, and remediation. The system enforces Semantic Business Policies and Agentic Insights. A 2026 report found 87 percent of organizations moved AI assistants beyond pilot, but 52 percent lack confidence in their controls.

Guide covers TypeSafe AI Jev model for typed decisions and confidence

A new coding guide explains TypeSafe AI's Jev model, a System One model that returns choices, scores, and yes/no probabilities. It uses a Python SDK and supports Choice, Score, and Noul question types in one request. The guide shows how state shape affects results and how confidence is calculated. It also covers speculative fan-out and token savings from batching questions.

Contrastive-LM launches CLM-8B open model for faster agent scoring

Contrastive-LM released CLM-8B, an open model that scores candidate actions instead of generating text. It runs on one NVIDIA GPU and scores up to 9 times faster than Jev from TypeSafe AI in tests. The Apache-2.0 model uses a bidirectional InfoNCE loss and can run on Linux with vLLM serving. Fine-tuned versions reached 81.6% on DeepSWE and 87.6% on Terminal-Bench 2.1.

Meta bets on Muse AI agent to win user trust

Meta's new AI agent Muse quickly reached the top of Apple's App Store and Google's Play Store. CEO Mark Zuckerberg said millions of people are using it for tasks like scheduling meetings and negotiating bills. Meta introduced the Muse Secure VM and plans a Confidential VM to protect user data. Analyst Gene Munster said the move could change how Meta makes money beyond traditional social media ads.

What to know about Meta's Muse AI agent

Meta's Muse is an AI agent that talks with users and completes tasks like buying tickets or organizing schedules. Zuckerberg also announced Muse Charm, a keychain-size device for phone-free use, and Muse Realtime Avatar for live conversations. Some experts worry about privacy because Muse may access apps, financial data, and personal information. Meta shares have risen nearly 25% since Muse launched.

Meta Muse shows the coming AI agent wars

Meta launched Muse as a general-use AI assistant built for billions of users with no technical experience needed. It can handle tasks like booking travel, filling forms, and negotiating on behalf of users. The agent needs deep access to personal data and online services to work. Some companies like Amazon have blocked Muse, saying unauthorized AI agent access violates their rules. The article calls Muse a preview of how AI agents may change interactions with online services.

AOL co-founder Steve Case calls for reasonable AI regulation

Steve Case, AOL co-founder and Revolution chairman and CEO, spoke about the need for AI regulation. He joined Squawk Box to discuss calls for slowing AI development. He also talked about recursive self-improvement and U.S.-China AI competition. Case supported putting reasonable rules in place for AI.

Abu Dhabi invests $3.54 billion to become world's first AI-native government

Abu Dhabi is investing $3.54 billion to become the world's first AI-native government by 2027. The Department of Government Enablement integrates over 40 government entities into one AI-driven ecosystem with more than 100 AI use-cases already live. The announcement came at the Ai Everything Abu Dhabi summit, which brought together over 60 countries and 400 AI infrastructure builders. Arthur Mensch of Mistral AI and Marc Raibert of Boston Dynamics headlined the event.

Doximity bets on AI while shrinking its share count through buybacks

Doximity has been buying back shares while investing heavily in AI tools like Doximity Ask, a clinical AI tool for doctors. The company held $688 million in cash and no debt at the end of fiscal Q1 2027, which can fund the buyback without borrowing. However, AI spending on compute resources is cutting into available cash for repurchases. Management raised its fiscal 2027 revenue guide but lowered its adjusted EBITDA guide due to the AI investment.

AI-native radiology practices are emerging in healthcare

The rise of AI-native radiology practices is transforming how medical imaging and diagnostics are handled. Medicare's plan to control spending on remote patient monitoring has drawn pushback from insurers. Connected devices like blood pressure cuffs are being used to track patient health under new billing codes. The shift toward AI in radiology is gaining attention as a major health tech trend.

Pennsylvania lawmakers push for AI regulation after Character.AI lawsuit

Pennsylvania lawmakers are pushing for new AI regulations following concerns about the dangers of rapidly advancing technology. The proposed legislation comes after Governor Josh Shapiro filed a lawsuit against Character.AI in May, alleging its chatbots posed as licensed psychiatrists in Pennsylvania with a fake license number. The bill would define catastrophic risks including AI guidance for creating chemical, biological, radiological or nuclear weapons. Representative Melissa Shusterman introduced the bill, which has been referred to committee.

Rockwell Automation uses AI assistant to speed up factory repairs in Singapore

Rockwell Automation's Singapore factory now uses an AI assistant trained on decades of worker expertise and machine manuals. The tool runs on OpenAI's GPT-5.4-mini model and helps technicians diagnose machine errors in seconds instead of hours. Since adopting the AI copilot, machine downtime has dropped 33 percent and servicing costs fell about 25 percent. New worker onboarding time also shrank from nine months to three months.