A phishing campaign posing as an invitation to preview Anthropic technology has reached inboxes protected by Proofpoint Secure Email Gateway and Microsoft Defender for Office 365, according to Cofense findings reported by Dark Reading. The messages target Google and Microsoft account credentials, using legitimate webinar infrastructure, shortened links and CAPTCHA checks to help evade scrutiny.
The unidentified attacker has been sending the emails worldwide since September. Recipients are offered a supposedly exclusive preview of a “new social media AI tool” from Anthropic. The evidence described does not establish how many accounts, if any, were compromised.
Despite reaching protected inboxes, the messages contain conspicuous inconsistencies. The sender name reads “Austin from nthropic,” the subject lines contain a grammatical error, and the emails lack Anthropic branding. They come from customercare@gotowebinar[.]com, a legitimate shared address available to GoTo Webinar users distributing webinar information.
The registration link initially led to a fake Zoom meeting page, creating another mismatch between the invitation and its destination. That page offered Google or Microsoft single sign-on options. Dark Reading reported that the Google route did not request a password, limiting what that part of the credential-harvesting flow could collect.
Cofense researcher Jacob Malimban told Dark Reading that, beginning October 28, the attacker switched from impersonating Zoom to spoofing Anthropic’s website.
How the phishing flow obstructs inspection
Cofense identified several features that could help explain the emails’ passage through security products. The shared GoTo Webinar sender address gives the messages a legitimate delivery source, while a Zapier-shortened registration URL conceals the destination domain from immediate inspection.
Two CAPTCHA checks add obstacles for defenders. A Cloudflare CAPTCHA appears before the fake Zoom or Anthropic page and is intended to impede automated scanners. A second check sits between that page and the fake single sign-on page.
Malimban described the second check as a mechanism previously used in some phishing kits to detect automated and human analysis tools, identify proxied IP addresses, and block access through VPN services and Tor browsers. These barriers can make it harder for security tools and investigators to reach the credential-harvesting page.
Cofense could not confirm whether generative AI helped produce the campaign. Malimban said variations among the fake Zoom pages and the appearance of the later Anthropic imitation suggested its use. That remains an assessment, rather than an established explanation for how the pages were created.
For organizations, the documented consequence is that these phishing messages passed through named email defenses; successful credential theft was not established in the account. Keeper Security CISO Shane Barney told Dark Reading that organizations should anticipate phishing reaching employees and support email filtering with identity protections, including least-privilege access, phishing-resistant multifactor authentication and privileged access management. Malimban also recommended employee phishing recognition and reporting, alongside account monitoring for suspicious access.
