A new cybersecurity report from Cofense details a sophisticated phishing campaign targeting enterprise users by exploiting their interest in artificial intelligence. The attackers are sending emails that appear to be exclusive invitations to a private preview of Anthropic's latest technology, capitalizing on the high demand for access to frontier AI models. While the emails promise a focused conversation on AI systems and research, the true objective is to harvest login credentials for Google and Microsoft accounts.
The campaign leverages the branding of GoTo Webinar, a widely used video teleconferencing platform whose legitimate meeting invitations are often not blocked by security tools. The fraudulent emails appear to come from an address resembling customercare@gotowebinar[.]com and feature official GoTo Webinar logos. However, researchers noted a subtle but critical inconsistency: the sender's name is listed as "Austin from nthropic," with the initial "A" missing from Anthropic. Despite this misspelling, the text contains no other obvious errors, making the deception more convincing.
When recipients click the link to register for the fake webinar, they are directed through a series of deceptive steps designed to evade security scanners. The attackers use URL shorteners and subdomains with dashes to hide malicious addresses, such as goo[.]gle-access-eventregister[.]icu for Google and office[.]com-access-sharepoint[.]com for Microsoft. These links initially lead to a Cloudflare CAPTCHA page, which serves as a checkpoint to block automated security tools while appearing legitimate to human users. After completing the first CAPTCHA, victims are taken to a page spoofing a Zoom meeting invitation, where they must enter their work email address.
Upon selecting options like "Continue with Google" or "Continue with Microsoft," a second CAPTCHA appears before presenting the final login prompt. This dual-CAPTCHA mechanism effectively prevents analysis by VPN services, Tor browsers, and other security scanners. The spoofed Microsoft page successfully captures both usernames and passwords, while the fake Google page currently only collects usernames, likely due to a misconfiguration that attackers may correct in future iterations. Once credentials are submitted, there is no confirmation message or redirection to legitimate vendor pages; instead, the data is immediately exfiltrated by the threat actors.
Frank Dickson of Dickson Research highlighted that the attackers did not choose Anthropic at random but selected the hottest topic in the current news cycle to exploit the "fear of missing out." He noted that the AI industry's reliance on gated previews and waitlists inadvertently trains employees to expect such exclusive invitations. The campaign demonstrates how threat actors use trusted tools and mimicry to defeat defenses built on trust, often combining three different brands—GoTo Webinar, Zoom, and Google or Microsoft—within a single attack chain.
Experts advise that relying solely on employee vigilance is insufficient, especially when attackers utilize AI to create highly convincing replicas of legitimate pages. Organizations are urged to verify any AI preview invitations through known contacts or account teams and to treat inconsistent branding across emails, meeting pages, and login screens as a major red flag. Furthermore, deploying phishing-resistant multi-factor authentication, such as passkeys, is recommended to render stolen passwords less valuable. If credentials are compromised, immediate password resets and activity reviews for the affected accounts are essential steps to mitigate further infiltration.
Sources: Computerworld.