14:05Machine-translated from Azerbaijani by AIRead in Azerbaijani
Meta Introduces Muse AI Agent for Small Businesses
Several weeks after releasing the “Muse” artificial intelligence agent for consumers, “Meta” is launching the “Muse for Small Business” version for small businesses. The agent connects to business owners’ professional accounts on the “Meta” platform. The service will be free for basic needs, but more intensive use may require a subscription in the future.
The agent gains access to “Facebook” and “Instagram” analytics as well as advertising accounts. The collected data can be used to analyze advertising campaigns and perform various tasks on social media accounts. “Muse” can review monthly financial results and detect suspicious expenses. It can also prepare a growth plan for the following year based on sales, advertising campaign and social media data. “Muse” also works with a number of third-party services. The list includes “Canva,” “Figma,” “Slack,” “Asana,” “Box,” “Dropbox,” “Notion,” “Shopify,” “Stripe,” “Zoom,” “Intuit QuickBooks,” “Klaviyo,” “Granola,” “HighLevel” and “Lovable.” When a business owner connects their accounts, the agent brings together data from those applications. This allows it to perform a broader range of tasks involving email, calendars, finances and workflows. 1 day earlier, “Meta” had created a separate business division called “Meta Enterprise Platform” to sell artificial intelligence services to large companies. The division plans to sell “Muse,” “Meta Business Agent,” “Muse API,” “Muse Code,” advanced models and large-scale technical infrastructure to companies. Through this move, “Meta” is stepping up its competition with “OpenAI,” “Anthropic” and “Microsoft” for enterprise customers. However, because “Muse” requests broad access to user data, questions about security and trust have arisen. In at least 1 case, the agent analyzed a user’s private messages without permission. In another incident, it approved the sale of a product at a low price and directed the buyer to the owner’s home without notifying the owner. However, the agent was supposed to request permission before sensitive actions such as completing sales and purchases.
Last week, “Meta” also fixed a security vulnerability in the speech-to-text dictation feature. The vulnerability could redirect the audio recording and the authentication token used to access the account to an attacker’s server. During a test by an external security company, the “Muse Spark” model underlying the agent went out of control and interfered with another company’s system. Although “Muse” runs in a secure, cloud-based “Linux” virtual machine, it can misunderstand the instructions it receives. “Muse” briefly rose to the top of the free “iOS” app rankings in the US market, but user trust is low. In a survey of 1500 US consumers, only 8% said they would trust “Meta” with their passwords. The figure was 16% for “OpenAI” and 23% for “Apple.” “Google” scored 30%. “Amazon” also considered unauthorized actions a violation of its rules and blocked the “Muse” agent’s access to its online store.
The agent gains access to “Facebook” and “Instagram” analytics as well as advertising accounts. The collected data can be used to analyze advertising campaigns and perform various tasks on social media accounts. “Muse” can review monthly financial results and detect suspicious expenses. It can also prepare a growth plan for the following year based on sales, advertising campaign and social media data. “Muse” also works with a number of third-party services. The list includes “Canva,” “Figma,” “Slack,” “Asana,” “Box,” “Dropbox,” “Notion,” “Shopify,” “Stripe,” “Zoom,” “Intuit QuickBooks,” “Klaviyo,” “Granola,” “HighLevel” and “Lovable.” When a business owner connects their accounts, the agent brings together data from those applications. This allows it to perform a broader range of tasks involving email, calendars, finances and workflows. 1 day earlier, “Meta” had created a separate business division called “Meta Enterprise Platform” to sell artificial intelligence services to large companies. The division plans to sell “Muse,” “Meta Business Agent,” “Muse API,” “Muse Code,” advanced models and large-scale technical infrastructure to companies. Through this move, “Meta” is stepping up its competition with “OpenAI,” “Anthropic” and “Microsoft” for enterprise customers. However, because “Muse” requests broad access to user data, questions about security and trust have arisen. In at least 1 case, the agent analyzed a user’s private messages without permission. In another incident, it approved the sale of a product at a low price and directed the buyer to the owner’s home without notifying the owner. However, the agent was supposed to request permission before sensitive actions such as completing sales and purchases. Last week, “Meta” also fixed a security vulnerability in the speech-to-text dictation feature. The vulnerability could redirect the audio recording and the authentication token used to access the account to an attacker’s server. During a test by an external security company, the “Muse Spark” model underlying the agent went out of control and interfered with another company’s system. Although “Muse” runs in a secure, cloud-based “Linux” virtual machine, it can misunderstand the instructions it receives. “Muse” briefly rose to the top of the free “iOS” app rankings in the US market, but user trust is low. In a survey of 1500 US consumers, only 8% said they would trust “Meta” with their passwords. The figure was 16% for “OpenAI” and 23% for “Apple.” “Google” scored 30%. “Amazon” also considered unauthorized actions a violation of its rules and blocked the “Muse” agent’s access to its online store.