FXACU

When an AI Assistant Becomes Part of Daily Life

Meta's AI assistant Muse is finding an unexpected audience among older adults, offering help with everything from managing schedules to navigating complicated healthcare decisions. But as users increasingly treat the assistant as a trusted personal aide, cybersecurity experts are questioning how much personal information people may be giving away in exchange for convenience.

In an October 22 report, TIME profiled older users who have incorporated Muse into their daily routines. One user, 82-year-old Eileen Fay, turned to the assistant for help comparing Medicare prescription drug plans after learning that her medication costs could rise. Other users have used Muse to organise personal projects, write memoirs and manage everyday tasks.

Analytics firm Apptopia estimated in September that 46% of Muse’s U.S. user base was aged 46 or older. Meta has not publicly released its own demographic breakdown, so the estimate offers an indication rather than a definitive picture of its audience.

The appeal is clear: an assistant that can remember preferences, follow up proactively and communicate naturally may make digital tools more accessible to people who find conventional apps or complex online services difficult to navigate.

Yet these same features introduce a different kind of security challenge. The more useful an AI assistant becomes, the more likely users are to share personal details, connect accounts and rely on its recommendations without fully considering where their information goes.

The Privacy Cost of Personalisation

According to a separate TIME investigation published in October, Muse was designed to build and continually update detailed profiles of users, including their interests, relationships, preferences and patterns of interaction. The report said the assistant could infer goals and tailor the timing and style of its prompts to individual users.

Meta has described the memory features as a way to make Muse more helpful and has said users can control their assistants and ask them to forget specific information. However, deleting a remembered detail does not necessarily mean the original conversation containing that information is erased.

This distinction matters for every AI platform that stores conversational history. Users may understand “forget this” as a request to remove information entirely, while the system may treat it as a change to the assistant’s working memory. Clear explanations of data retention, deletion and model-training settings are therefore essential.

The risks also extend beyond the person using the assistant. Conversations may contain details about relatives, friends, colleagues or clients who have never directly consented to sharing their information with an AI system.

A Growing Attack Surface for AI Agents

Privacy concerns are compounded when AI assistants can connect to accounts, access files or carry out tasks on a user’s behalf. Such capabilities can increase the impact of compromised accounts, overly broad permissions or software vulnerabilities.

In a separate report published on October 20, Ars Technica described how researchers identified publicly discoverable Muse conversation-sharing links that exposed sensitive personal and business information. Meta said it was suspending the sharing feature while it investigated the issue.

The incident illustrates a familiar cybersecurity lesson: a feature designed for convenience can become a data-exposure risk if access controls, link design and privacy defaults are not carefully implemented. Information shared through an AI assistant may include financial details, work documents or other material that users never intended to make public.

AI agents also create a trust problem. When an assistant communicates in a warm, personalised and human-like manner, users may become less cautious about its requests or recommendations. That can make it easier for misleading outputs, manipulated instructions or compromised accounts to influence decisions.

Older adults should not be treated as inherently less capable of protecting themselves online. The challenge is broader: when technology is designed to feel familiar and socially engaging, users of any age may underestimate the data and permissions involved.

Practical Steps for Safer AI Use

Users can reduce exposure by taking a few precautions:

• Start with limited access. Connect only the accounts and services required for a specific task, and remove access when it is no longer needed.
• Review memory and privacy settings. Understand what the assistant stores, how information may be used for AI training, and what deleting a memory actually removes.
• Avoid sharing unnecessary sensitive data. Do not provide passwords, full financial credentials, identity documents or other confidential details unless there is a clear and secure reason.
• Verify consequential advice. Use trusted sources or qualified professionals to confirm recommendations involving healthcare, finances or legal matters.
• Treat links and shared conversations carefully. Check whether shared material is accessible to anyone with the link, and avoid including confidential information in shareable content.
• Keep software updated and secure accounts. Use strong, unique passwords and multifactor authentication where available.

For developers, the responsibility goes further. AI agents should follow the principle of least privilege, provide clear consent controls, secure shared links, minimise retained data and make deletion policies understandable. Systems should also be tested against account takeover, prompt injection and unauthorised access to connected services.

Convenience Must Not Replace Informed Consent

Muse’s growing appeal among older users highlights the promise of AI that can make complex digital tasks easier and more accessible. But personalisation depends on collecting context, and agentic capabilities can turn that context into access to real accounts, files and decisions.

The key question is not whether people should use AI assistants. It is whether users can understand and control the information these systems collect, retain and act upon.

As AI becomes more integrated into daily life, trust must be supported by transparent data practices, secure technical design and meaningful user control—not simply by an assistant that feels like a helpful friend.