
Infostealer malware no longer needs your users' passwords. Copying their session cookies is enough to skip login and MFA entirely. Google's answer, Device Bound Session Credentials (DBSC), is now available in Chrome 146 on Windows. It ties a session to a private key held in the device's TPM, so stolen cookies stop working within minutes. Unlike most browser security features, this one needs changes on your server.
What happened
In April 2026, BleepingComputer reported that Google had added DBSC protection in Chrome 146 for Windows, with macOS (using the Secure Enclave) promised in a later release. Google first announced DBSC in 2024. It developed the protocol with Microsoft as an open web standard, with specifications published through the W3C.
Google says it tested an early version for about a year with several web platforms, including Okta, and saw a notable drop in session theft. It also acknowledges that software alone can't stop cookie theft once malware is on a machine, because the malware can read the files and memory where browsers keep cookies. DBSC is designed so that what the malware copies quickly stops being useful.
Who is affected
- Chrome 146+ users on Windows with a TPM, for sites that have implemented DBSC.
- Web developers and identity teams who want to protect long-lived sessions, especially for admin consoles, SaaS dashboards, finance apps and developer platforms that infostealers target.
- Not yet covered: macOS, other browsers, and any site that hasn't added the registration and refresh endpoints.
What it means for developers and engineering teams
DBSC replaces one long-lived bearer cookie with a short-lived cookie plus a proof-of-possession key:
- After login, your server sends a
Secure-Session-Registrationheader listing supported algorithms (for example ES256, RS256) and a registration path such as/StartSession. - Chrome generates a key pair in the TPM. The private key can't be exported. Chrome then posts the public key, inside a JWT, to that endpoint.
- Your server binds the key to the session and returns a JSON config (
session_identifier,refresh_url,scopeandcredentialsnaming the cookie). It then swaps the long-lived cookie for a short-lived one, for example with a 10-minute max-age. - When the cookie expires, Chrome holds the pending request and calls
refresh_url. The server responds 403 with aSecure-Session-Challenge. Chrome signs it and returns aSecure-Session-ResponseJWT, and the server issues a fresh cookie.
An infostealer can still copy the current cookie, but it expires within minutes and can't be refreshed without the TPM key. Each session uses its own key, and only that public key is shared, so DBSC doesn't add a cross-site device identifier.
Some practical caveats. DBSC works only on HTTPS and doesn't support Partitioned cookies. Chrome may send requests without the short-lived cookie if the refresh endpoint is unreachable, the TPM is busy, or third-party cookies are blocked. Keep a fallback path, and decide deliberately what an unbound request may do. A good pattern is to allow low-risk reads but require re-authentication for sensitive actions.
What to do now
- Read Chrome's DBSC developer guide and the W3C spec, and prototype the
/StartSessionand refresh endpoints in your auth service or middleware. - Start with high-value sessions: admin panels, billing, and API-token and SSH-key management pages.
- Shorten cookie lifetimes for DBSC-bound sessions, and log refresh failures so you can tell TPM problems from real attacks.
- Keep server-side defences for non-DBSC browsers: session binding heuristics (IP or ASN changes, user-agent drift), re-authentication for sensitive operations and fast session revocation.
- Check with your identity provider (Okta, Entra, Auth0, Clerk and others) about DBSC support before building it yourself.
- Keep treating infostealers as an endpoint problem too: EDR, no unmanaged browsers for admin work, and rotating any tokens found in stealer logs.