Kief Studio

Cross-Platform Agentic Browsing: Managing Security Across Devices

Protect

  • Published
  • Updated
  • 8 min read
A bright ribbon of violet and gold light enters from the left and fans into many fine strands on black

TL;DR: Cross-platform agentic browsing means an AI agent that follows you from laptop to phone to home computer, carrying your sessions and context along. The risk is the weakest device in that chain. Sort devices by trust, keep regulated work off unmanaged ones, use phishing-resistant sign-in, set session timeouts, and know how to cut a lost device off.

Agentic browsing used to live on one machine. Now the same agent, with the same signed-in accounts and memory of what you were doing, can pick up on your phone where your work laptop left off. That is genuinely useful. It also means a task you started on a managed office computer can finish on a family tablet nobody has updated in a year.

We first published this note in 2025 with a set of device statistics and named case studies we could not trace to a real source, so they are gone. What follows sticks to what NIST, CISA, OWASP and the browser makers publish, plus the practical setup I would put in place for a small team before letting an agent work across their devices.

What is cross-platform agentic browsing, and why does it widen the risk?

Control room with screens showing purple signal waves, like one agent session seen from many devices

Cross-platform agentic browsing is an AI agent in the browser that reads pages, fills forms and takes actions for you, with its accounts, history and context available on every device you sign in to.

What actually travels between devices

Ordinary browser sync already moves a lot. Google's help page says that when you sign in to Chrome you "can get your bookmarks, passwords, and more on all your devices." (Google Chrome Help (opens in a new tab)) Agentic browsing adds the agent's view of what you were working on and the sessions it uses to act. Before you turn it on, find out exactly what your agent syncs and where that data is stored. The vendor's own documentation should say.

Why every device becomes a door

Think of it as handing one assistant a key to every office you use. The office with the weakest lock is now the way in to all of them. A personal phone without a passcode, a shared home computer, a laptop that stopped getting updates: if any of them is signed in to the same agent, it can reach what the best-protected machine can reach.

Where does agentic browsing leak data across devices?

Neon purple chip glowing on a dark circuit board

Most of the risk in cross-platform agentic browsing is not a clever attack. It is ordinary sync doing exactly what it was built to do, in a place nobody planned for.

Sync that ignores your boundaries

A client file opened on the office laptop shows up in the agent's history on a personal phone. A research session about a confidential deal is waiting on the home computer the next morning. Nobody broke in. The data simply moved to a device your policies never covered. As an example, picture a three-person accounting office in Grafton where one partner reviews client returns at home on a shared family laptop. If the agent syncs there, so does the work.

Shared and family devices

Shared devices are where agentic browsing goes wrong most quietly. A spouse or a teenager opens the browser and the agent is still signed in, with your sessions. Chrome's own guidance is to "add a profile in Chrome" if you share your computer with others. (Google Chrome Help (opens in a new tab)) For business use, I would go further: no business agent on a shared device at all.

Forgotten devices

Old phones in a drawer, a laptop handed back by a former employee, a tablet used once at a trade show. If they are still signed in, agentic browsing still works on them for whoever picks them up. My employee offboarding checklist for IT access (opens in a new tab) covers the account side; the device side is the same discipline.

How does prompt injection travel across devices?

Shield with a padlock at the center of streams of blue data

An agent reads content written by strangers and then acts on it. OWASP describes the risk: "Indirect prompt injections occur when an LLM accepts input from external sources, such as websites or files." (OWASP Gen AI Security Project, LLM01 Prompt Injection (opens in a new tab)) OWASP also says "it is unclear if there are fool-proof methods of prevention for prompt injection."

In cross-platform agentic browsing, a page the agent read on one device can shape what it does later on another, if its memory or context syncs. That is a reason to keep agent memory narrow, to keep confirmations on for sending, buying, deleting and sharing on every device, and to give the agent the fewest accounts it needs. We go deeper on the attack in prompt injection attacks explained.

What does zero trust say about personal devices?

Web of purple and orange nodes joined by thin lines, like devices linked to one account

It says not to trust a device because of where it sits or who owns it. NIST's definition of zero trust: it "assumes there is no implicit trust granted to assets or user accounts based solely on their physical or network location (i.e., local area networks versus the internet) or based on asset ownership (enterprise or personally owned)." (NIST SP 800-207, Zero Trust Architecture (opens in a new tab))

For agentic browsing that is the right starting point. Being signed in on one trusted machine should not make every other device trusted. Each device should prove what it is and earn access to each system on its own.

NIST also notes that mobile devices "were initially personal consumer communication devices but they are now permanent fixtures in enterprises and are used to access modern networks and systems to process sensitive data." (NIST SP 800-124 Rev. 2 (opens in a new tab)) That guidance on managing mobile devices is the practical companion to zero trust for anyone letting agentic browsing run on phones.

How should you sort devices for agentic browsing?

Orange data tunnel rushing toward a dark center

The original version of this note proposed trust tiers, and they still hold up. Sort every device that touches business accounts into one of four groups and let agentic browsing do only what each group allows.

  1. Managed company devices. Issued, updated and monitored by the business. Full agentic browsing for business work, with logging.
  2. Enrolled personal devices. Personal phones or laptops enrolled in device management, with a passcode, encryption and current updates required. Agent use for everyday business tasks, not regulated or client-confidential data.
  3. Unmanaged personal devices. Basic protection, no company control. Reading and light research only. No sync of client data.
  4. Shared, borrowed or unknown devices. No business agent at all.

Write the four groups on one page, list which devices sit in each, and review the list every quarter. That list is most of what a small team needs.

Which controls matter most for agentic browsing across devices?

Teal code rain with floating purple shards on a dark background

Phishing-resistant sign-in

CISA's guidance is blunt: "The use of MFA on your accounts makes you 99% less likely to be hacked." (CISA, Multifactor Authentication (opens in a new tab)) NIST's newest authentication guideline goes further and says applications assessed at its middle assurance level "must offer a phishing-resistant authentication" option. (NIST SP 800-63B-4 (opens in a new tab)) Passkeys and hardware security keys are the usual way to get there. Amelia, my wife and Kief Studio's co-founder and CEO, explains MFA fatigue (opens in a new tab) and why push approvals alone are not enough.

Session timeouts

An agent that stays signed in forever on every device is the forgotten-device problem in another form. NIST says an overall reauthentication timeout "SHOULD be no more than 24 hours at AAL2" and "The inactivity timeout SHOULD be no more than 1 hour." (NIST SP 800-63B-4 (opens in a new tab)) Those numbers are written for federal systems, but they are a sensible default for agentic browsing sessions on business accounts.

Separate profiles and narrow accounts

Give agentic browsing its own browser profile, signed in only to the accounts the agent needs. Keep banking, payroll and admin consoles in a different profile that agentic browsing never touches. Treat the agent as its own identity with its own permissions; she covers this in what a non-human identity is (opens in a new tab).

What should you do when a device is lost or compromised?

Neon cyan crystal towers rising over dark water

Speed matters more than perfection. A short, written incident response list for cross-platform agentic browsing might read:

  1. Sign the device out everywhere. Revoke its sessions from the account settings of the browser, the agent and every business app it touched.
  2. Change the passwords and keys the agent could reach, starting with email.
  3. Wipe or lock the device through device management if it was enrolled.
  4. Read the agentic browsing activity log for the last few days. Note anything sent, bought, shared or deleted.
  5. Decide who needs to know. If client or patient data may be involved, talk to your attorney about notice duties.

Practice this once on a test device. The first time you do it should not be the day it matters.

What can a small team do this week?

Glowing city of purple and cyan circuit paths seen from above

  • List every device signed in to a business account or an AI agent, including old ones.
  • Sign out the ones you no longer use.
  • Check what your agent syncs and turn off sync for history and memory on anything outside the first two groups.
  • Turn on phishing-resistant sign-in for email first, then every account the agent can use.
  • Set session timeouts where your tools allow it.

That is an afternoon of work, and it removes most of the risk that cross-platform agentic browsing adds.

Glowing blue and purple shield etched with circuit lines on black

Frequently asked questions

Is cross-platform agentic browsing safe for business use?

It can be, on devices you manage, with phishing-resistant sign-in, session timeouts and confirmations on risky actions. It is not safe on shared or unmanaged devices that hold the same sessions.

Should employees use agentic browsing on personal phones?

Only on phones enrolled in device management with a passcode, encryption and current updates, and only for everyday tasks. Keep regulated or client-confidential work on company devices.

Does zero trust apply to AI agents?

Yes. NIST's zero trust model grants no trust based on location or ownership. Each device and each agent identity should be checked before it reaches a system.

What should I turn off first?

Sync of agent history and memory to any device you do not manage, and the business agent on any shared or family computer.

How long should an agent stay signed in?

NIST suggests no more than 24 hours overall and 1 hour of inactivity at its middle assurance level. That is a sensible default for business agent sessions.

This article is general information, not legal advice. Kief Studio is not a law firm, and reading it does not create an attorney-client relationship. Laws and regulations change, so check the current text and talk with a licensed Massachusetts attorney about your situation.

Where we fit

Orange interface panels and charts floating on a dark grid

Kief Studio is in Shrewsbury, and we help teams across Massachusetts put structure around tools like this. Our zero trust architecture work maps identities, devices, applications and the paths between them, puts identity checks and least privilege in front of each system, and rolls changes out in stages so the team keeps working while access tightens. If you are planning to roll out agents and want the design checked first, a security architecture review maps what each tool and vendor can reach and recommends changes in order of risk.

First conversation is free. No commitment.

Tell us which devices your team works from

Related articles

  1. Protect

    Identity and Access Management for AI Browser Agents

    Identity and access management for AI browser agents: give each agent its own identity, scope it to one task, require approvals and log who acted for whom.

    Brian Gagne

  2. Protect

    Zero Trust for Worcester Medical Offices: Where a Small Practice Starts

    Zero trust for a small Worcester medical practice: what it means, how it maps to the HIPAA Security Rule, and a staged plan that keeps patient care running.

    Brian Gagne

  3. Protect

    AI Browser Performance Optimization: Speed vs. Security Trade-offs

    Why an AI browser slows down, what security actually costs in speed, which controls to keep, and how to measure and tune performance on real office machines.

    Brian Gagne