{
  "id": 6374662,
  "title": "Meta Muse vs ChatGPT Agent: Two Companies Built Two Very Different Cages for AI That Acts Like You",
  "url": "https://urgent.news/2026/09/09/meta-muse-vs-chatgpt-agent-two-companies-built-two-very-different",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-09T03:11:35.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/jamilxt/meta-muse-vs-chatgpt-agent-two-companies-built-two-very-different-cages-for-ai-that-acts-like-you-1gg8"
  },
  "original_language": "en",
  "account": "On September 8, Meta unveiled an AI agent called Muse, which granted the agent access to users' email, calendar, and payment capabilities. This launch came shortly after OpenAI faced scrutiny for allegedly stealing a math proof, which may have played a role in Meta's timing. My own AI agent infrastructure, which handles tasks such as article publishing and content management, features strict boundaries, including no access to bank accounts, email, or purchases. I thoroughly examined the architecture documents of both companies, leading to this analysis.\n\nA crucial distinction between the two AI agents lies in their safety measures. Muse, available exclusively in the US, functions within a virtual machine dedicated to each user, with its own browser. Untrusted web data is isolated, while the action-performing component remains separate, never exposed to sensitive information. Sentinel, a separate process, scrutinizes every outbound action, ensuring compliance with user-approved policies. For sensitive operations like sending email or making purchases, Sentinel escalates to a human approval process, preventing unauthorized actions even if the underlying model is compromised.\n\nPayments are handled through Stripe's Link, which generates a single-use card number for each transaction, preventing the agent from accessing actual card details. Meta's safeguards extend to user privacy, allowing users to opt out of their interactions being used to train Meta's models and to request deletion of specific data. A \"Confidential VM\" upgrade is in the works, isolating the entire VM within a trusted execution environment, with access keys stored on the user's device. Meta is partnering with Moxie Marlinspike, Signal's founder, to further enhance security and has committed to a substantial bug bounty program to encourage reporting of vulnerabilities.",
  "summary": "On September 8, Meta handed a consumer AI agent three things that used to be considered untouchable: access to your email, your calendar, and a way to spend your money. The agent is called Muse, it launched the same day OpenAI was fielding accusations about a stolen math proof, and the timing was probably not an accident. A launch that would normally dominate the tech news cycle got buried under…",
  "key_points": [
    "Meta unveils AI agent Muse with email, calendar, payment access",
    "OpenAI faces scrutiny over alleged theft of math proof",
    "Muse operates in isolated virtual machines with Sentinel safety"
  ],
  "editors_take": "Meta's expansive approach to AI agent access, despite robust safety measures, contrasts with more restrictive architectures, indicating a diverging industry trend on balancing capability and security.",
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}