A phishing campaign impersonating Anthropic’s artificial-intelligence services has bypassed secure email gateways from Proofpoint and Microsoft, according to research from Cofense. Active since September, the campaign uses an invitation to a supposedly private Anthropic AI preview to target Google and Microsoft account credentials.
The emails claim to offer an exclusive, invitation-only preview of a new Anthropic social-media AI tool. They appear to come from the legitimate shared GoTo Webinar address customercare@gotowebinar[.]com, although the sender name contains a typo and the message links to neither Anthropic nor GoTo Webinar. Cofense said it observed the messages reaching environments protected by Proofpoint Secure Email Gateway and Microsoft Defender for Office 365.
The campaign’s effectiveness comes from layering legitimate infrastructure and anti-analysis measures rather than from sophisticated malware. The registration link appears to use Zapier’s URL-shortening service, while the email abuses GoTo Webinar branding to make the sender look routine. Cofense said the link first leads to a Cloudflare CAPTCHA, then to a fake Zoom event page. A second, sliding CAPTCHA appears before the victim reaches counterfeit Google or Microsoft single-sign-on pages; the additional checkpoint can hinder automated scanners and traffic routed through VPN or Tor services.
The phishing flow is inconsistent and contains obvious warning signs. The invitation’s GoTo Webinar branding changes to Zoom, while later versions observed beginning Oct. 28 reportedly spoofed Anthropic instead. Cofense said the fake Google page appeared to collect usernames but not passwords, likely because of a configuration error, whereas the Microsoft-themed page collected both. The researchers could not confirm that generative AI produced the campaign, although variations among the spoofed pages were consistent with possible AI-assisted development.
No CVE is associated with the incident, and Cofense has not attributed it to a named threat actor. The episode instead highlights a continuing limitation of reputation- and content-based email filtering: a message sent through a trusted service can reach the inbox even when its destination is malicious. Organizations should treat unexpected AI-preview invitations as phishing, verify offers through Anthropic or a known business contact, inspect changes in branding and domains, and require phishing-resistant multifactor authentication. Users should also navigate directly to official services rather than signing in through links delivered in unsolicited email.
