The fastest way to get an AI agent to do real work is also the worst one: give it your password. This week, one of the biggest password managers shipped a way around that, and it is a good moment to fix how your team handles logins before agents arrive in force.

What launched

On 22 October 2026, 1Password released 1Password for Agents. It lets you allow an AI agent to sign in to your accounts without the agent ever holding your credentials.

The flow is simple:

  1. The agent tries to log in and sends 1Password a unique digital key to prove which agent it is.
  2. You get an alert about the login attempt and confirm it.
  3. 1Password fills in the credentials on the agent's behalf. The agent gets in, but never sees the password.

At launch it works with Anthropic's Claude Duet, OpenAI's Dots and ChatGPT Work, Grok Bot, Meta's Muse and agents from the startup Town. 1Password says more are coming. In its own words, handing over a password is "a common shortcut to an agent", and this is meant to replace it.

It builds on an earlier feature. In July, 1Password added an Agentic Mode to its browser extension that locks the vault while an agent is in control. The agent can only use the logins approved for the current task, you approve with biometrics, and access ends when the task is done.

Why this matters now

Agents are moving from demos into daily work. On the same day, Uber put all 30,000 of its employees on ChatGPT Work. Agents that can read your inbox, update a CRM record or download an invoice need to sign in somewhere, and today most of them do it with a shared password pasted into a chat.

Banks have noticed. According to an Axios report cited by TechCrunch, US banks have started detecting agent activity in customer accounts, and most cannot tell whether the account owner asked for it. Chase's advice to customers is blunt: do not give your username and password to anyone. An agent counts.

The other risk is plain leakage. A password typed into an agent's chat can end up in logs, in a shared workspace or in a vendor's systems. The same week, developer platform Railway asked all of its users to revoke their secret keys after a cyberattack. Every shared secret is one more thing to rotate when something goes wrong.

What this changes for your CRM and back office

Most small and mid-sized teams still run on a handful of shared logins: one CRM admin account, one supplier portal login, one bank user for the finance team. That was already a problem for accountability. With agents it gets worse, because you lose the ability to say who, or what, made a change.

The good news is that the fix is the same one that makes a CRM trustworthy anyway:

  • One user per person and one per agent. Your CRM's audit trail is only useful if every action has a named owner. An agent should appear as "Lead follow-up agent", not as "Admin".
  • The smallest access that does the job. An agent that books follow-up calls does not need to export your contact list or change deal values. Give it a role with only the fields and actions it needs.
  • Approve, then log. For sensitive steps, such as sending a quote, changing a payment detail or deleting records, keep a human approval in the loop and record it.

Five rules to set before you let an agent log in

  1. Stop sharing passwords in chat, with people or with agents. Move the team to a password manager first. It is the foundation everything else sits on.
  2. Give every agent its own account or API key. Where a system offers an API, use it instead of letting an agent click through the website with a person's login.
  3. Scope it to one job. List what the agent is allowed to read, create and change. If you cannot write that list, the agent is not ready.
  4. Keep humans on the irreversible steps. Payments, deletions and anything sent to a customer should wait for a click from a person until the agent has a track record.
  5. Review the log weekly. Check what each agent did, what failed and what it was refused. That review is where you find the next thing worth automating.

The bottom line

Tools like 1Password for Agents remove the worst habit, which is pasting a password into a chatbot. They do not decide what an agent should be allowed to do inside your business. That part is a process and permissions question, and it is worth answering before the agents arrive, not after the first surprise in your audit log.

If your CRM still runs on shared logins, start there. Named users, clear roles and an audit trail are what make it safe to let AI agents do the busywork.