New CSS-Based Attacks Can Break Webmail Isolation to Steal Passwords and Tokens

Security researcher Gareth (PortSwigger) has disclosed a set of new attack techniques that use CSS to break the isolation between rendered email content and the surrounding webmail interface. According to the report, published by The Hacker News, content inside an email can "escape" its message boundary and interfere with the webmail client itself, rather than staying confined to the message body.

What was reported

The research describes attack chains affecting multiple major webmail providers, including Outlook, Gmail, Fastmail, Proton Mail, Yahoo Mail, and AOL Mail. Depending on the target, the techniques are reported to be capable of capturing passwords, taking over connected third-party accounts, leaking authentication tokens, hijacking trusted UI actions within the mail client, and manipulating AI tools that read or summarize email content.

Why it matters for defenders

Webmail interfaces sandbox untrusted email content specifically so that a malicious message can't reach into the surrounding application chrome or session context. If CSS alone is sufficient to break that boundary across several independent, widely used providers, it suggests a systemic weakness in how HTML/CSS email rendering is isolated rather than a single vendor bug. Any organization relying on these webmail platforms — and any user of AI-assisted email tools that parse message content — is potentially in scope. Because the vector is CSS rather than JavaScript, it may also slip past defenses tuned primarily to block script-based email threats.

What defenders should watch for now

• Treat this as a phishing/credential-theft delivery risk: monitor for unusual or unexpected webmail login prompts, password re-entry requests, or OAuth/token consent screens surfaced immediately after opening an email.
• Review email gateway and webmail provider security bulletins for patches or mitigations related to CSS rendering in the message pane.
• Hunt for anomalous UI-driven actions (e.g., automated clicks on "trusted" buttons, unexpected account-linking or token-grant events) that could indicate hijacked webmail UI actions rather than user-initiated activity.
• If your organization uses AI email assistants/summarizers, evaluate whether they render or process raw HTML/CSS from messages, and consider stricter sanitization or disabling automatic content parsing until vendor guidance is available.
• Encourage users to be cautious with password prompts or account-linking requests that appear while an email client is open, especially if unsolicited.

This is developing, net-new intelligence — full technical details, proof-of-concept specifics, and per-vendor patch status were not fully covered in the available summary. Defenders should follow the original research for updates: The Hacker News: New CSS Attacks Can Break Webmail Defenses to Steal Passwords and Tokens.