Account takeover
Email account takeover is when an attacker signs in to a mailbox as its real owner. The way in is a stolen password, a phished session or a reset the attacker triggers. Once inside, the attacker reads mail, resets other accounts through it, hides behind inbox rules and writes to contacts as the owner.
Updated
How account takeover works
Account takeover works by getting the mail service to accept the attacker as the owner. Codes and resets, built to protect the account, are often the route in.
- The attacker gets a way to sign in. A password is phished, reused from a breach elsewhere, or guessed.
- If a one-time code is required, the attacker gets that too. A fake sign-in page relays the code as it is typed, or keeps the session the real site hands back.
- Or the attacker skips the password. A reset link or a sign-in link goes to a mailbox or phone the attacker can already read.
- The attacker signs in. To the mail service this is the owner.
- The attacker makes the access last: a forwarding rule, a new sign-in method, a connected app. These can outlive a password change.
- The attacker uses the mailbox: searches it for invoices and credentials, resets other accounts that send their codes here, and replies inside real threads.
What account takeover looks like
An account takeover rarely looks like anything to the owner. The first visible sign is usually a trace the attacker left. In the mailbox of Megan Park at Trotta, it could be a rule like this.
Rule name: .
If: subject contains "invoice" or "payment"
Then: forward to archive@example.net
mark as read, move to RSS SubscriptionsThe rule copies the valuable mail out and hides it from Megan. Microsoft lists such rules among the common symptoms of a compromised mailbox.
Signs of account takeover
The signs of account takeover are changes in the mailbox that its owner did not make.
| Sign | What it means |
|---|---|
| Inbox rules or forwarding you did not create | Mail is being copied out or hidden. Remove them and assume a takeover. |
| Sign-in alerts from places or devices you do not know | The password or a session is in other hands. |
| Codes or reset messages you did not request | Someone is trying the reset route, here or on an account tied to this mailbox. |
| A new sign-in method, recovery address or connected app | The attacker is building a way back in. |
Why account takeover gets through ordinary filtering
Account takeover gets through because nothing is forged. The attacker signs in with valid credentials, so the mail service sees its own customer. Mail the attacker then sends comes from the real mailbox and passes SPF, DKIM and DMARC at every receiver.
One-time codes do not close the gap alone. CISA notes that some forms of multi-factor sign-in can be phished, and names FIDO sign-in as phishing resistant. Gmail and Outlook screen incoming mail and warn about unusual sign-ins.
When an AI agent reads the mail
An AI agent with access to a mailbox is a second way to take over the accounts behind it. A mailbox receives sign-in codes, reset links and sign-in links, and an agent that can read them can complete a reset. An attacker who hijacks the agent with an injected instruction does not need the password at all.
An agent's API key is also a credential. A key that can read and send is a signed-in session with no second factor.
An agent that receives mail from a taken-over account sees a known sender and passing authentication. Nothing tells it that the writer changed.
How to stop account takeover
Account takeover is stopped by making a stolen password or code useless, and by keeping codes and resets away from anything that can be tricked.
General controls
- Use phishing-resistant sign-in, such as passkeys or security keys, on every mailbox.
- Do not type a code into a page you reached from a message, and do not read one out to a caller.
- After any suspicion, do more than change the password: end all sessions, and remove unknown rules, forwarding, sign-in methods and connected apps. Microsoft's guidance lists each step.
- Give agents narrow keys, rotate them, and keep codes and resets out of their reach.
At the door, with a pre-delivery inbox
A pre-delivery inbox narrows two common routes. Trotta Mail screens inbound mail while the sending server is still connected. In the sign-up and verification use case, a fake sign-in message is held outside the inbox for a human.
Login mail in a human's inbox starts sealed. A one-time code, a password reset link or a sign-in link there does not exist for any API key, whatever its scopes, until a signed-in human opens it to keys.
An agent's key cannot read held mail or loosen protection, and it cannot send from a human's inbox unless a signed-in human allows it. Humans can turn on two-factor sign-in, and the audit log records sign-ins and key changes. See the security model.
No filter catches everything. When a mailbox elsewhere is taken over, mail from it is real mail from a real account, and a second channel is what tells you.
Questions
Does changing my password end an account takeover?
Not always. Forwarding rules, added sign-in methods, app passwords and connected apps can outlive a password change. End all sessions and remove anything you did not set up.
Can attackers get past two-factor authentication?
They can get past codes and push prompts by relaying them through a fake sign-in page, or by stealing the session after sign-in. Passkeys and security keys resist this because they answer the real site alone.
How do attackers use password resets to take over accounts?
Most services send a reset link or code to an email address. Whoever can read that mailbox can reset the account. This is why one taken-over mailbox leads to many more accounts.