The risk in this phishing campaign is bigger than a fake Anthropic invitation. It turns familiar enterprise services into a chain of credibility: a webinar email, a registration page and a Google or Microsoft sign-in. By the time recipients reach the credential prompt, each step can feel like a routine part of joining an online event.
Cybersecurity company Cofense describes attackers using this sequence to collect work account details. The lure is a private, invitation-only session about AI systems and frontier model research, aimed at employees eager to learn what Anthropic might be developing. The pitch taps into a real workplace pressure: teams are trying to understand where AI fits, and access to a limited preview can sound like a useful professional opportunity.
Frank Dickson of Dickson Research says the appeal is deliberate. “It sells exclusivity,” he said. “Fear of missing out is the oldest lure in the book.” Anthropic is the timely name in this campaign, but Dickson expects the brand to change as the news cycle shifts. “The attackers did not pick Anthropic at random,” he said. “They picked whatever is hottest in the news cycle. Next month, it may be another name.”
A familiar webinar invite conceals the first redirect
The campaign borrows GoTo Webinar’s identity, a practical choice because organizations commonly receive invitations from the platform and security controls may allow those messages through. Cofense found that the email appears to come from a legitimate sender and directs recipients to register. The link instead leads through a compromised URL toward the attackers’ credential collection pages.
The invitation promises a “focused conversation on AI systems, research, and the future of frontier models” for a select group. Its text has no glaring spelling mistakes, and it carries GoTo Webinar branding. The sender address appears as customercare@gotowebinar[.]com, although GoTo Webinar’s website is www.goto.com. A more revealing detail is the sender name: “Austin from nthropic,” with the first letter of Anthropic missing.
That small error sits alongside a more consequential mismatch in the journey. The initial email presents a GoTo Webinar event, while the registration experience imitates Zoom. Later, recipients encounter Google or Microsoft sign-in pages. Dickson calls this a sequence of three brands in three clicks, and says the changing identities offer useful clues for anyone who slows down to inspect the flow.
The registration link adds another layer of concealment. Cofense says a URL shortener helps bypass email security tools, while subdomains and hyphens make the destination harder to read at a glance. The link appears to have been generated by Zapier, whose tools can connect with Claude, ChatGPT and other systems compatible with Model Context Protocol (MCP). The use of a recognizable service name can make a link look more plausible, even when its destination deserves scrutiny.
CAPTCHAs help the operators screen out analysis
After a recipient follows the link, a Cloudflare CAPTCHA appears. Completing it leads to the Zoom-style event page, where the visitor is asked to enter a work email address and choose “Continue with Google” or “Continue with Microsoft.” A second CAPTCHA then appears before the login page.
CAPTCHAs are common on legitimate sites, so their presence can reassure a person that a page is using an ordinary security measure. In this campaign, however, Cofense says the two challenges also appear to block analysis by VPN services, scanners, Tor browsers and other security tools. Requiring separate challenges along the route may make it harder for automated systems to follow the whole sequence.
“A CAPTCHA was built to keep bots out,” Dickson said. “Here, it keeps the security scanners out and lets the victims in.” That reversal matters for security teams: a challenge that looks routine to an employee may be serving the operators by filtering the visitors most likely to expose the site.
The final pages imitate Google and Microsoft sign-in screens closely. Their domains are crafted to look familiar when skimmed. The Google page uses goo[.]gle-access-eventregister[.]icu; the Microsoft page uses office[.]com-access-sharepoint[.]com, combining recognizable service terms in a misleading address.
The two forms do not behave identically. Cofense found that the fake Google page collects usernames but not passwords, which researchers believe may reflect a configuration mistake. The Microsoft page collects both. That difference does not make the Google prompt harmless: a username can still be useful to an attacker, and the researchers expect the collection setup could be corrected.
Credential theft can become an account compromise
Neither imitation login page provides a confirmation message or sends the visitor on to the legitimate vendor site after credentials are submitted. By then, the operators have already received the information entered into the form. Depending on what they capture and what protections are in place, a stolen account could expose sensitive documents or provide an initial route deeper into an enterprise, Cofense warns.
The immediate defense is to treat the full path as evidence. Check whether the branding stays consistent from the original email through the event page and login prompt. Look for small discrepancies such as “nthropic,” and inspect the actual domain before entering credentials. A familiar logo or polished layout is not proof that a sign-in page belongs to Google or Microsoft; the address matters.
Cofense also cautions that attackers can reproduce the appearance of legitimate services, with AI making convincing imitation easier. Visual checks still help, but they cannot carry the whole defense. As Dickson puts it, “Training helps, but I would not hang the defense on an employee squinting at a URL at 4:45 on a Friday.”
Make stolen passwords less useful
The industry’s use of waitlists and gated previews can train employees to expect exclusive invitations, Dickson argues. That context helps explain why the lure works: the email resembles a familiar kind of access offer, and the recipient may be acting in good faith to learn about a technology relevant to their work.
Organizations can reduce reliance on individual vigilance by confirming preview invitations through an account team or another known contact, and by treating changes in branding across an email, meeting page and login screen as a warning. Phishing-resistant multi-factor authentication, including passkeys, can also make a stolen password far less useful.
If someone has entered credentials on one of these pages, Dickson recommends resetting them and reviewing the account’s activity. The campaign’s central lesson is practical: trusted tools and familiar branding can be assembled into a convincing route to a fake login. Enterprise defenses need to account for that route before an employee is asked to spot every flaw in the moment.