Real Google MFA prompts still finished under hostile observation while attackers kept the Workspace session. Operation CloudFire, reported around 4 November 2026, steered Google Calendar invites and later Drive shares through Google redirectors into a Nyasher reverse proxy sitting on genuine Google sign-in. Victims entered passwords and completed SMS codes, authenticator codes, Tap-Yes number matching, or phone recovery while the proxy captured authenticated cookies and tokens. Phishing-proof, device-bound authentication removes those transferable factors at the login ceremony, so that reverse-proxy path never completes and never issues a Google Workspace session to the operator. Sessions already copied after a finished legacy login still need revoke and OAuth cleanup; that residue is separate from stopping the proxied sign-in.
How CloudFire turned Calendar lures into live Workspace cookies
According to BleepingComputer, covering Material Security's Operation CloudFire work, the live path started mid-August 2026. Calendar invites carried voicemail, RFP, summons, and payment themes, often from previously compromised sender addresses people already trusted. Links rode Google /url redirectors and related Google-hosted hops, later shifting toward Drive file and document shares, with share.google paths and one Intuit-flavored variant in the mix.
Those hops landed on JavaScript-built capture pages fronted by Cloudflare Turnstile. Behind the checks sat Nyasher, a reverse-proxy framework documented publicly since February after a hardcoded key appeared in its JavaScript. The proxy presented real Google sign-in chrome and stayed in the middle of the conversation. Material tracked the wave from the start, later reporting detections that caught over 99% of malicious messages in its customer base with a 0.01% reclassification rate. Late September 2026 is when Material stood up CloudFire and fed gateway and backend URLs to Cloudflare abuse. Over roughly a month of pressure the counters reached 68,676 blocked messages, 2,112 unique malicious URLs analyzed, 897 domains taken down, and 8,570 unique compromised sender addresses across 6,354 root domains. Public reporting does not name victim organizations or publish exact stolen-session counts.
An adversary-in-the-middle attack places a live reverse proxy between the user and the real identity provider so password and second-factor traffic can be relayed and the issued session collected. CloudFire's Nyasher instance did exactly that against Google sign-in. BleepingComputer's summary of Material's findings states the human moment cleanly: the victim enters a password, receives a real prompt, and approves it, all while the operator watches the session in real time.
The server relayed SMS codes, authenticator codes, Google's Tap-Yes number matching, and phone-based account recovery to Google as each response arrived. None of those second-factor controls stopped the session from being captured. Every factor on that path is something a person can type, read aloud, or tap while a middlebox copies the outcome. The origin the user believed they were binding to was not the origin that kept the cookies.
| Factor on the path | Proxy behavior | Outcome |
|---|---|---|
| Password | Relayed live to Google | Login continued |
| SMS or authenticator code | Relayed as entered | Second factor cleared |
| Tap-Yes number matching | Relayed approval | Prompt satisfied |
| Phone account recovery | Relayed recovery step | Account still unlocked |
| Issued Workspace session | Cookies and tokens copied | Mail, Drive, OAuth ready |
After the cookies left the proxy, operators held live Workspace sessions rather than a reusable password alone. Material's framing is direct: when the outcome is a hijacked session rather than a stolen password, the evidence appears in Workspace audit logs. Defenders who hunt only for impossible-travel password events miss the signal. From that point the work is post-sign-in containment: revoke sessions, review OAuth grants issued in the user name, reset credentials, and treat mail and Drive as exposed until proven clean. Shortening cookie lifetime after the fact does not unwind a session the attacker already holds. An unfinished Microsoft 365 phishing branch sat in the same kit but did not resolve at reporting time. The confirmed damage stayed on Google Workspace identities. For the full attack-chain narrative, see how CloudFire stole Google Workspace sessions.
Stopping the proxied Google login before cookies issue
Device-bound, origin-bound authentication changes the ceremony the proxy depends on. A phishing-proof enterprise IdP issues only hardware-bound signatures. The private key never leaves the device TPM or secure enclave, never crosses the network as a typed secret, and never appears as a push the user can approve on a hostile page. The signature is bound to the legitimate origin. A reverse proxy that presents a lookalike capture page cannot complete that proof, because there is no SMS code, authenticator OTP, Tap-Yes approval, or recovery secret to relay.
On CloudFire's initial-access path that matters immediately. The Nyasher kit only wins if the user finishes a transferable login under observation. Remove the transferable factor and the kit never receives authenticated Google Workspace cookies from that path. Prevention, not detection is the point: the attack fails before a session exists for the operator, rather than hoping gateway filters catch every Calendar lure after the fact.
Once a live session cookie already sits with the operator, login controls have nothing left to challenge. Mail access, Drive reads, and OAuth grants in the victim name are post-authentication use of a stolen session. Revoke, OAuth app review, Conditional Access signals, and endpoint hygiene handle that residue. Phishing-proof MFA does not erase cookies already copied from a completed proxied login. It removes the cheap remote path: live reverse-proxy completion of Google sign-in with relayable factors.
| Stage | What attackers held | Device-bound phishing-proof MFA |
|---|---|---|
| Proxied Google login | Password plus relayed MFA | Login cannot complete remotely |
| Captured Workspace cookies | Live mail, Drive, OAuth | Does not un-steal issued tokens |
| Later endpoint malware | Local session after real login | Separate residual problem |
That split is why detection and takedown still mattered operationally. Nearly nine hundred domains burned and tens of thousands of messages blocked shortened the campaign's runway. Prevention at the ceremony is a different layer. Operators even staged unfinished Microsoft 365 phishing paths beside the live Google code, which shows how portable the same transferable-factor model remains across IdPs when the second factor can still be relayed.
Passkeys, recovery, and full-lifecycle coverage
Passkeys and FIDO2-style WebAuthn place a key pair on the device, keep the private key in hardware, and produce an origin-bound signature at authentication. Against a reverse proxy that only works when secrets and approvals move across the wire, that design can stop the relay at the login act. On a generic enterprise IdP that issues only hardware-bound signatures, the capture page is the wrong origin and has no OTP or push to steal.
Passkeys harden authentication. They do not automatically harden every earlier or later lifecycle step unless the deployment forbids phishable enrollment and recovery. CloudFire's public record does not publish passkey outcome data, so the comparison stays architectural: an origin-bound login can block the relayable ceremony; it is still weaker than full-lifecycle phishing-proof coverage when recovery remains a typed channel.
CloudFire did not stop at password plus authenticator. Phone-based account recovery was explicitly in the relay set. A support-style or self-service recovery flow that still accepts SMS, voice, or other transferable proofs gives the same middlebox another chance to finish enrollment or unlock under the attacker's eye. Once an attacker enrolls their own factor during a window of access, cookie lifetime limits stop mattering. They hold a legitimate proof.
MFA 2.0 is phishing-proof across registration, device onboarding, authorisation, authentication, and decommissioning. Adding a device requires an already-enrolled device, not a recovery code a proxy can read. The enrolment model is part of the control, not an afterthought bolted onto a strong login. That is the strength ladder in practice: password is weaker than OTP and SMS, which are weaker than push, which is weaker than origin-bound login such as a passkey, which is weaker than phishing-proof coverage across the whole identity lifecycle.
Key takeaways for defenders
- Require device-bound, origin-bound authentication on Google Workspace workforce logins so reverse proxies cannot relay a second factor.
- Treat phone recovery and other transferable unlock paths as part of the same failure class as OTP and Tap-Yes on a proxied page.
- Hunt Workspace audit logs for hijacked-session and unexpected OAuth-grant evidence, not only stolen-password events.
- Revoke sessions and kill OAuth apps fast when a CloudFire-style cookie theft is confirmed; that work is containment after the login already failed.
- Assume unfinished cross-IdP branches exist in the same kits and extend the same no-transferable-factor policy beyond a single cloud suite.
Workforce Google Calendar lures will keep borrowing trusted senders and Google redirectors. The durable fix is a login and recovery design with nothing left for a reverse proxy to copy.
FAQ
Would phishing-proof MFA have stopped Operation CloudFire's Google login path?
Phishing-proof MFA would have stopped Operation CloudFire's Google login path by refusing transferable factors a Nyasher reverse proxy can relay. Device-bound, origin-bound signatures never send SMS codes, authenticator OTPs, Tap-Yes approvals, or recovery secrets across the network, so the proxied ceremony cannot complete and Workspace session cookies are never issued to the operator on that path. Cookies already stolen after a completed legacy login still need revoke and OAuth cleanup; that post-sign-in work is not undone by changing the login factor.
Why did standard Google MFA fail against the CloudFire reverse proxy?
Standard Google MFA failed against the CloudFire reverse proxy because SMS codes, authenticator codes, Tap-Yes number matching, and phone recovery are all transferable proofs a live middlebox can forward. According to Material Security reporting via BleepingComputer, victims finished real prompts while the operator watched and captured the authenticated session. The second factor cleared on Google's side and the cookies still went to the attacker.
Are passkeys enough to stop a CloudFire-style Workspace AiTM kit?
Passkeys can block a reverse proxy that depends on relayable secrets at the Google sign-in ceremony, but they are not automatically phishing-proof across enrollment and recovery. CloudFire also relayed phone-based account recovery, which remains a gap if passkey deployments still allow phishable unlock or device-add flows. Full-lifecycle phishing-proof MFA closes registration, onboarding, authentication, and decommissioning without a typed recovery secret.
What should security teams do after a CloudFire-style hijacked Workspace session?
Security teams should revoke Google Workspace sessions immediately, review and remove OAuth apps granted in the victim name, reset credentials, and search Workspace audit logs for session and grant activity rather than only password anomalies. Material Security and BleepingComputer both stress that hijacked-session evidence lives in those logs once MFA already succeeded under the proxy. Shortening token lifetime after theft does not remove access the attacker already holds.
Did Operation CloudFire also attack Microsoft 365 with the same kit?
Operation CloudFire did not activate a live Microsoft 365 branch in the disclosed campaign. Operators shipped unfinished Microsoft 365 phishing code, including stylesheet paths and a Microsoft sign-in title string, beside the live Google paths, but those routes did not resolve when researchers reported the work. Confirmed session hijacks ran against Google Workspace sign-in and Workspace apps.