top of page

Human or AI Agent? How Businesses Can Tell Who’s Really Logging In

4 hours ago
12 min read


Human or AI Agent?

How Businesses Can Tell Who’s Really Logging In


Human or AI Agent? How to Spot, Block, and Stop AI Agents


Your next customer, or the next login on your company email, might not be a person. Here’s how organizations can spot, welcome, block, and stop AI agents, with an action plan for executives, CIOs, CISOs, and consumers.


By Theresa Payton, CEO of Fortalice Solutions and former White House Chief Information Officer  |  November 12, 2026



Human Agents
Human Agents

On a Monday morning, the CIO of a 2,000-employee regional distributor gets two messages before her first cup of coffee.


The first is from the e-commerce team. A longtime customer’s account placed six orders overnight, changed its shipping address, and logged in from a cloud data center three states away. Is it fraud? Or is it the customer’s new AI agent doing exactly what it was asked? Nobody can tell.


The second is from the security team. A routine permissions review found 37 employees who had connected personal AI agents to their work email, including two in accounts payable. No one approved them. Every action those agents took shows up in the logs under the employees’ own names.


Neither problem required a hacker. By Friday, she had a plan for both, and it’s the same plan laid out in this post.


This scenario is a composite, not a real client. But every detail reflects capabilities and risks that exist today.


For most of the internet’s history, one simple assumption held: if someone logged in, a person was on the other end. That assumption no longer holds. Personal AI agents such as Meta Muse, OpenAI dots, xAI’s Grok Bot, and Instinct can open websites, log in, fill out forms, send email, and make purchases. Many run in the vendor’s cloud, use a real web browser, and move at human speed. To a website, a well-built agent can look exactly like you.

Here’s the good news: the tools to tell the difference are arriving fast. The industry is converging on a simple, sensible idea. A trustworthy agent should show ID, and a business should be able to decide what each agent may do.


Part One: Customers and Their AI Agents

Why the old warning signs stopped working

Traditional bot defenses look for things software does badly: clicking too fast, failing a CAPTCHA puzzle, or announcing itself with a robotic label. Today’s agents use real browsers, move at human speed, and can solve many puzzles. Most importantly, when a customer hands an agent their username and password, the agent is logged in as that customer. That’s exactly why the distributor’s e-commerce team couldn’t tell fraud from a helpful agent.



Key Takeaways

•     2.3% of legitimate AI agent sessions presented a verified digital ID in the third quarter of 2026, according to Arkose Labs. Verification is the right direction, but it’s still rare.

•     88% of employees who have an AI tool from work also use personal AI tools for work tasks, according to Gartner. Personal agents are already inside most organizations.

•     3 ways agents show up: with a name tag, with a verified signature, or with no ID at all.

•     The goal is a front door with a guest list, not a locked building. Many businesses can turn on agent verification through providers they already use.



Three ways an agent shows up at your door

•     With a name tag. Some agents label their web traffic. Muse, Grok Bot, and Google’s agent do this. It’s helpful, but anyone can write a name tag, so on its own it proves very little.

•     With a verified signature. A newer standard called Web Bot Auth lets an agent digitally sign each request, a bit like a notarized letter. The business checks the signature against a public key the agent’s maker publishes. OpenAI’s agents and Grok Bot sign their requests, and Google is testing it. A forger can’t fake a valid signature without the maker’s private key.

•     With no ID at all. Some agents, including Instinct and agents running on a person’s own laptop, don’t identify themselves. They look like ordinary customers.


How to spot an agent

1.   Check for a valid signature. Cloudflare, Akamai, HUMAN, AWS, and Vercel can verify Web Bot Auth signatures automatically. For many businesses, this is a setting to turn on, not a project to build.

2.   Check the name tag against where the traffic comes from. A request claiming to be a known agent but arriving from the wrong network is a red flag for impersonation.

3.   Watch behavior. How an account moves, what it buys, how fast it changes settings, and where it logs in from still matter. Banks have been spotting agents that log in with customers’ passwords this way.


How to welcome the trustworthy AI Agents

•     Approved data connections. Banks such as JPMorgan Chase allow read-only sharing through open banking providers while blocking agents that log in with a customer’s password. The customer gets convenience, and the bank keeps control of anything that moves money.

•     Payment network registration. Visa’s Trusted Agent Protocol and Mastercard’s Agent Pay let merchants recognize registered shopping agents and tell them apart from malicious bots. Mastercard says only registered agents can transact on Agent Pay, using special payment tokens that mark a purchase as agent-initiated.

•     Rules with limits. Walmart lets compliant outside agents buy within limits the customer sets, such as $50 or $100, and blocks agents that don’t follow its rules.


How to block

•     Update your terms. United Airlines, eBay, and Amazon have all written agent restrictions into their customer terms. Clear terms give a business grounds to cancel agent-made transactions.

•     Block password-sharing logins. This is what most large banks have done. Agents using a customer’s password are turned away, while approved connections stay open.

•     Set rules at the edge. Your web security provider can allow, slow down, challenge, or block traffic by agent type. That turns a technical scramble into a policy choice.


How to stop an agent mid-task

•     Bring the human back for high-risk steps. Ask for a fresh login, a passkey, or a one-time code before moving money, changing a password, or updating a shipping address. Well-built agents are designed to hand these steps back to the person.

•     Keep the ability to reverse. Holding periods, confirmation emails, and easy cancellation give everyone a chance to catch mistakes. United’s new terms let it cancel agent-made reservations.



AI Agents
AI Agents

Part Two: Employees and Their AI Agents

How personal agents get into the workplace

No IT rollout is needed. An employee signs up for a personal agent, connects it to work email, the company drive, or a chat workspace, and clicks “approve” on a standard permission screen. From that moment, the agent has ongoing access to whatever that employee can reach. Because the agent runs in the vendor’s cloud rather than on the employee’s laptop, many traditional security tools never see it.

The Cloud Security Alliance has a name for this: a “shadow agent,” meaning an agent acting with an employee’s work access that the organization hasn’t approved, inventoried, or monitored. The 37 employees in our opening story weren’t being reckless. They were trying to get their work done faster.

Why this is hard to see

Company logs were built to answer one question: which account did this? Until now, that also told you who did it. When an agent uses an employee’s access, every log entry shows the employee. The employee was allowed to read that mailbox. Nobody approved the agent.

Early user reports show what to watch for: an agent pulling a verification code from an inbox to finish a sign-up, an agent still summarizing email after the user thought they’d disconnected it, and a login followed by a two-factor prompt from an unfamiliar overseas location.

How to spot, block, and stop shadow agents

•     Review app permissions. Microsoft 365 and Google Workspace admin tools show which outside apps employees have granted access to email, files, and messaging. Look specifically for agent vendors.

•     Use the security tools you already have. Cloud access security tools can flag personal accounts and unapproved AI services touching company data. Some endpoint security products now detect desktop agents such as Muse.

•     Look for logins from vendor clouds. Activity on an employee’s account that comes from an AI vendor’s data center, rather than the employee’s usual device and location, is a useful signal.

•     Require approval for sensitive permissions, separate personal accounts from corporate ones, and know how to revoke an agent’s access without locking out the employee.

Offer a better path

Blocking alone tends to push work onto personal devices, where security teams see even less. The more durable approach is a governed alternative. Enterprise versions of these agents are arriving. OpenAI, for example, is letting businesses set up agents with their own identities and credentials for tasks like procurement and invoicing, so the agent is visible and accountable. When the approved option is good, the shadow option loses its appeal.


Other Considerations

•     A signature proves who built the agent, not who sent it. Web Bot Auth confirms an agent came from OpenAI or Google, for example. It doesn’t confirm which customer sent it or what they actually wanted. Payment network programs from Visa and Mastercard are designed to close that second gap.

•     Disconnecting isn’t always deleting. At least one user reported an agent still working after access was removed. Confirm revocation; don’t assume it.

•     Anyone who can email your employee can talk to their agent. An agent reading a work inbox can be steered by instructions hidden in an incoming message. For the two accounts payable employees in our story, that turns an ordinary phishing email into a way to direct a payment.

•     Blocking is a pause, not a strategy. The banks, airlines, and retailers blocking agents today are also building or joining sanctioned paths.


Your Action Plan: Prepared and Resilient

Here is what the CIO in our fictionalized story did that week, broken out by role, plus what each of us can do as customers and employees.


Business executives: questions to ask your team this quarter

1.   Would we know if an employee connected a personal agent to work accounts tomorrow?

2.   Can we tell the difference in our logs between an employee and an agent using that employee’s access?

3.   What is our strategy for customers who shop through agents: block, welcome, or welcome with limits?

4.   Who owns agent policy across IT, security, legal, and the business, and when will it be written down?

5.   Does offboarding remove agents, not just people?

6.   Could we reconstruct what an agent accessed and sent if a customer or regulator asked?


CIOs: Your Checklist

•     Turn on agent verification. Ask your web security or content delivery provider to enable Web Bot Auth checks and agent labeling.

•     Bring payments into the conversation. Evaluate Visa’s Trusted Agent Protocol and Mastercard’s Agent Pay with your payments and e-commerce teams.

•     Require admin approval for outside apps that request access to mail, files, or messaging in Microsoft 365 or Google Workspace.

•     Pilot a governed enterprise agent so employees have an approved option that’s better than the shadow one.

•     Add agent permissions to offboarding and role changes, alongside accounts, devices, and badges.


CISOs: Your Checklist

•     Set edge rules by agent type. Allow verified agents, challenge unverified ones, and block anything impersonating a known agent.

•     Require step-up verification for payments, password resets, and changes to bank details, contact information, or shipping addresses.

•     Hunt for shadow agents. Review third-party app grants quarterly, and look for employee logins from AI vendor data centers.

•     Test revocation. When you disconnect an agent, confirm it has actually stopped.

•     Treat inbound email as an agent attack surface. Update phishing training and incident response playbooks to cover hidden instructions aimed at agents, especially in finance and accounts payable.

•     Make sure your logs can reconstruct agent activity, including what was accessed, sent, and changed.


Consumers and employees: Your Checklist

•     Welcome the extra check. If your bank or a retailer asks you to verify again before a payment or address change, that’s the system making sure you are still in charge.

•     Check your company’s policy before connecting any AI agent to work email or files.

•     Be careful with the inbox that gets your codes. An agent that can read the email where your verification codes and password resets arrive can act with a lot of your power.

•     Review your connected apps every few months. When you disconnect an agent, remove it in two places: inside the agent’s app and in your account’s security settings. Then confirm it has stopped.

•     Prefer agents that identify themselves and use registered payment options when they’re available.

Your Questions, Answered

Can a website tell if I’m using an AI agent?

Sometimes, and it’s getting better at it. Some agents label themselves, some carry a verified digital signature through a standard called Web Bot Auth, and some carry no ID at all. Websites also watch behavior, such as login location and how quickly settings change. In the third quarter of 2026, only 2.3% of legitimate agent sessions carried a verified signature, so behavior still does much of the work.

What is Web Bot Auth?

A digital ID card for AI agents. It lets an agent sign each request with a cryptographic key. The website checks that signature against a public key published by the agent’s maker, much like checking a notary’s seal. Cloudflare, Akamai, HUMAN, AWS, and Vercel can verify these signatures automatically.

How do Visa and Mastercard verify AI shopping agents?

Through registration programs built on Web Bot Auth. Visa’s Trusted Agent Protocol and Mastercard’s Agent Pay help merchants recognize registered shopping agents and separate them from malicious bots. Mastercard uses special payment tokens that mark a purchase as agent-initiated, and says only registered agents can transact on Agent Pay.

What is a shadow agent?

An AI agent using an employee’s work access without company approval. The term comes from the Cloud Security Alliance. It usually happens when an employee connects a personal agent to work email, files, or chat with a single click. Because the agent runs in the vendor’s cloud, many security tools can’t see it.

Can my employer see if I connect an AI agent to my work email?

Often, yes. Microsoft 365 and Google Workspace admin tools show which outside apps have been granted access to company accounts, and many security tools flag unapproved AI services. The best move is to check your company’s policy before connecting anything.

Why does my bank or airline keep asking me to verify again?

To make sure a human is still in charge of the big decisions. Extra checks before payments, password changes, or contact updates are one of the most effective ways to stop an agent, or a criminal, from acting without you. Well-built agents are designed to hand those steps back to you.

How do I fully disconnect an AI agent from my accounts?

Remove it in two places, then confirm. Disconnect the account inside the agent’s app, then revoke its access in your email or account security settings under connected or third-party apps. If you ever shared a password with the agent, change it. Then check that the agent has truly stopped, since at least one user has reported an agent still working after disconnecting.

Should businesses block AI agents or welcome them?

Both, deliberately. Block agents that log in with shared passwords or ignore your terms. Welcome verified, registered agents through controlled paths with spending limits and human approval for high-risk steps. Customers will increasingly expect to use agents, and the businesses that build a safe front door will earn that trust.

What is the first thing a business should do this week?

Two quick checks. Ask your web security provider whether agent verification is available to turn on, and ask IT to review which outside apps employees have connected to company email and files. Both usually use tools you already own.

Who can help my company detect and manage AI agents?

A cybersecurity firm that works across identity, fraud, and AI. Fortalice Solutions helps organizations find shadow agents, set agent policies, test customer-facing defenses, update incident response playbooks, and train employees. Learn more at www.fortalicesolutions.com.

Need Help? Fortalice Solutions Is Here

At Fortalice Solutions, our promise is fortified security: making sure our clients are prepared for an attack and resilient to future threats. Telling humans from agents, and good agents from bad ones, is now part of that work.




Need Help? Fortalice Solutions Is Here

At Fortalice Solutions, our promise is fortified security: making sure our clients are prepared for an attack and resilient to future threats. Telling humans from agents, and good agents from bad ones, is now part of that work.

•     For businesses: we can help you find shadow agents already connected to your systems, test whether your website and apps can tell customers from agents, write a practical agent policy, update your incident response plans, and train your teams, especially in finance and accounts payable.

•     For executives and their families: we help protect the personal accounts, inboxes, and digital footprint that agents and fraudsters both target.



Ready to build a front door with a guest list? 

Visit www.fortalicesolutions.com or call 877.487.8160 to talk with our team.


About the Author

Theresa Payton is the CEO of Fortalice Solutions and served as the first female White House Chief Information Officer. She began her career in banking technology, security, and fraud prevention, and today advises boards and executives on cybersecurity and AI. She is the author of “Manipulated: Inside the Cyberwar to Hijack Elections and Distort the Truth.”

T

his post summarizes publicly available reporting and vendor documentation as of November 11, 2026. Product capabilities and standards are evolving quickly; confirm current details with your providers. The opening scenario is a composite created for illustration.


Sources

Primary and standards sources

•     Cloudflare, Securing agentic commerce with Visa and Mastercard: https://blog.cloudflare.com/secure-agentic-commerce

•     Cloud Security Alliance, Grok Bot, Dots, Muse and Instinct now act with your employees’ access: https://cloudsecurityalliance.org/blog/2026/11/10/grok-bot-dots-muse-and-instinct-now-act-with-your-employees-access

•     Axios, Visa Trusted Agent Protocol: https://www.axios.com/2025/10/14/visa-ai-shopping-agent-protocol-bot

Analysis and reporting

•     Centinel Analytica, What a website can verify about AI browser agents: https://www.centinelanalytica.com/blog/ai-browser-agents

•     Lume, How to detect AI agents (Arkose Labs Q3 2026 figure): https://lumetraffic.com/blog/detect-ai-agents-on-your-website

•     AI Crawler Check, How personal agents identify themselves: https://aicrawlercheck.com/blog/personal-ai-agents-website-guide

•     Eco, Mastercard Agent Pay vs. Visa Trusted Agent Protocol: https://eco.com/support/en/articles/15192003-mastercard-agent-pay-vs-visa-trusted-agent-2026-compared

•     The Hacker News, Bring Your Own Agent: https://thehackernews.com/2026/10/bring-your-own-agent-your-employees.html

•     Aembit, Personal AI agents and enterprise secrets: https://aembit.io/blog/blog-personal-ai-agents-enterprise-security/

•     NeuralTrust, Meta Muse security for CISOs: https://neuraltrust.ai/blog/meta-muse-ai-agent-security

•     Harmonic Security, Shadow AI detection: https://www.harmonic.security/solutions/shadow-ai-detection

•     Forcepoint, What is shadow AI: https://www.forcepoint.com/blog/insights/what-is-shadow-ai

•     G3NR8, Gartner 88% personal AI use figure: https://www.g3nr8.com/blog/ai-is-getting-personal-muse-dots-duet

•     PPC Land, United Contract of Carriage Rule 6.J.6: https://ppc.land/united-bans-ai-agents-from-buying-its-tickets-seats-and-upgrades/

Fortalice Solutions

•     Fortalice Solutions: https://www.fortalicesolutions.com



Comments


Commenting on this post isn't available anymore. Contact the site owner for more info.
bottom of page