Deceptive links
A deceptive link is a link whose visible text names one destination while its real address opens another. In email, the text can show a trusted address, a button or a short link, and the address underneath can pass through redirects before it lands on a page built to collect a sign-in or deliver a file.
Updated
How a deceptive link works
A deceptive link works because a link in an HTML message has two independent parts: the text a reader sees and the address it opens.
- The attacker builds the destination: a copy of a sign-in page, or a page that offers a file.
- The attacker sets the link text to something trusted: a familiar address, a button that says View document, the name of a known service.
- The attacker sets the address to the harmful page, often indirectly. A link shortener hides it, or a redirect on a well-known site forwards to it.
- The message adds a reason to click now: a deadline, a locked account, a shared file.
- The reader clicks. The browser follows every hop without asking and shows the last page, which asks for a password or offers a download.
What a deceptive link looks like
A deceptive link looks like any other link, because the mismatch is in the markup. This one reads as the sign-in page of Trotta.
Link text: https://trotta.io/ login
Address: https://login.example.net/ trotta.io/ sessionThe real name appears in both lines. In the second it is a folder on someone else's site. With a shortener the address would be https://short.example/x7Qa, which says nothing about where it ends.
Signs of a deceptive link
The signs of a deceptive link are in the address, never in the text. Read the host name, the part between https:// and the next slash, from its right-hand end.
| Sign | What it means |
|---|---|
| Hovering shows a different address from the text | The two were set separately. Trust the address. |
| The known name comes after the first single slash | It is a folder on another site. The owner is named just before that slash. |
| The known name is followed by more dots and another domain | The owner is at the right-hand end. The known name is a label it chose. |
| A short link in a message about an account or a payment | The sender chose to hide the destination. |
| An @ sign inside the address | A browser can discard what comes before it. The real host comes after. |
Why deceptive links get through ordinary filtering
Deceptive links get through because a filter has to judge an address whose destination can change after delivery. The first hop is often a reputable shortener or a redirect on a legitimate site, so the link looks clean when it is scanned. The attacker can point it at the harmful page later.
A link also sidesteps the checks made on files: MITRE ATT&CK notes that attackers send links to avoid defenses that inspect attachments. Gmail and Outlook do filter spam, malware and known harmful links before delivery. A fresh address on a new domain has no history to judge.
CISA says more than 90% of successful cyberattacks start with a phishing email.
When an AI agent reads the mail
An AI agent does not hover over a link, so the oldest human check does not exist for it. Given plain text, an agent sees the link text or the address, not both. Given HTML, it sees both and may never compare them.
An agent with a browsing tool follows links fast and without doubt. If it holds credentials, it may type them into the page. Whatever the page says then becomes new input to the model, which is a route for prompt injection.
How to stop deceptive links
Deceptive links are stopped by judging the address, not the text, before anyone can follow it.
General controls
- Do not sign in from a link in a message. Open the service from a bookmark or type its address.
- Hover on a computer, or press and hold on a phone, and read the address before opening it.
- Use phishing-resistant sign-in such as passkeys or security keys. They answer the real site alone, so a copy collects nothing. CISA names FIDO sign-in as phishing resistant.
- Give a browsing agent a short list of domains it may open, and keep credentials away from it.
At the door, with a pre-delivery inbox
A pre-delivery inbox reads a message while the sending server is still connected. Trotta Mail takes every message apart before delivery, including where its links go. In the sign-up use case, a fake sign-in message whose link text and destination do not match is held outside the inbox for a human.
Each delivered message lists its links, each with a verdict and reasons. An agent's key reads the safe view, where links that are not clean are defanged so they cannot be followed by accident. See concepts and the security model.
No filter catches everything. A destination that turns harmful after delivery is hard for any screen to see, so the habits above still apply.
Questions
How can I check where a link goes without clicking it?
On a computer, hover over the link and read the address your mail client shows. On a phone, press and hold the link until the address appears. If it is a short link, treat the destination as unknown.
What is an open redirect?
An open redirect is a page on a legitimate site that forwards visitors to any address placed in the link. Attackers use it so the link begins with a trusted name and ends on their own page.
Is a link safe if the email passed SPF, DKIM and DMARC?
No. Authentication proves which domain sent the message. It says nothing about where the links go, and a compromised or lookalike sender passes.
Sources
- MITRE ATT&CK T1566.002, Phishing: Spearphishing Link
- Microsoft Security Blog, Widespread credential phishing campaign abuses open redirector links (August 2021)
- CISA, Implementing Phishing-Resistant MFA (fact sheet)
- CISA, Shields Up guidance: more than 90% of successful cyber-attacks start with a phishing email