In the rapidly evolving landscape of corporate technology, few names carry as much weight or generate as much excitement as Anthropic. As the developer of the Claude series of large language models, the company has positioned itself at the vanguard of the artificial intelligence revolution, drawing intense interest from developers, executives, and researchers alike. However, this high level of public and professional interest has not escaped the notice of cyber adversaries. Hackers are now gleefully exploiting the industry’s collective "fear of missing out" (FOMO) to launch a sophisticated phishing campaign designed to harvest sensitive corporate credentials.
New research from the cybersecurity firm Cofense has pulled back the curtain on a deceptive operation that hooks overeager enterprise users by offering them what appears to be a golden ticket: an invitation to a private, invite-only Anthropic preview event. By mimicking the branding of established communication platforms and leveraging the prestige of the AI industry, these adversarial actors are successfully tricking employees into surrendering their Google and Microsoft login credentials.
The campaign is a masterclass in psychological manipulation and technical evasion. According to Frank Dickson of Dickson Research, the success of the lure lies in its promise of exclusivity. "It sells exclusivity," Dickson noted. "Fear of missing out is the oldest lure in the book."
In a corporate environment where every organization is currently scrambling to determine where and how AI fits into its long-term strategy, a private preview of Anthropic’s newest frontier models is exactly the kind of invitation an ambitious or curious employee would be inclined to accept. The attackers’ choice of Anthropic as the bait was a calculated move. "The attackers did not pick Anthropic at random," Dickson explained. "They picked whatever is hottest in the news cycle. Next month, it may be another name."
How attackers lure users with spoofed Anthropic invites
The campaign begins with a highly polished email that leverages the reputation of GoTo Webinar, a widely used platform for corporate digital events. Because GoTo Webinar is a legitimate service frequently used for professional communications, its automated invitation emails are typically not flagged or blocked by standard email security tools. Cofense researchers discovered that attackers are exploiting this inherent trust by sending invitation emails that appear to originate from a legitimate sender.
The content of the email is crafted to be as enticing as possible. It describes a "tantalizing" opportunity, stating: "This private, invitation-only virtual session brings together a select group of participants to a focused conversation on AI systems, research, and the future of frontier models." For an employee working in tech, strategy, or operations, such an invitation feels like a significant career opportunity or a vital piece of competitive intelligence.
To bypass technical defenses, the attackers employ several layers of obfuscation. They utilize URL shorteners to hide the true destination of the links contained within the email, preventing security software from easily identifying malicious domains. Furthermore, the link itself appears to have been generated by Zapier, a popular automation tool. This is a particularly clever detail, as Zapier is frequently used by professionals to connect various AI tools like Claude and ChatGPT with other software via the Model Context Protocol (MCP). The presence of a Zapier-linked URL adds a layer of technical legitimacy that can disarm even skeptical users.
Despite the high level of polish, the researchers noted that the campaign is not entirely without flaws. The email sender’s address is listed as customercare@gotowebinar[.]com, whereas the actual GoTo Webinar website is located at www.goto.com. Perhaps more tellingly, the email identifies the sender as "Austin from nthropic," conspicuously missing the letter "A" from the company’s name. While the body of the email is free of obvious spelling or grammatical errors and features correct GoTo Webinar branding, this minor typo serves as a critical warning sign for those paying close attention.
Once a recipient clicks the phishing link, they are not taken directly to a registration page. Instead, they are met with a Cloudflare CAPTCHA page. While CAPTCHAs are traditionally used to ensure that a user is a human and not a bot, in this context, the tool is weaponized against the user. The CAPTCHA serves as a checkpoint to prevent automated security tools and web scanners from accessing and analyzing the malicious content on the attacker’s site.
"Because CAPTCHAs are common on legitimate websites, phishing recipients may believe all is well and everything is secure," the Cofense researchers explained. By mimicking the security protocols of a legitimate site, the attackers build a false sense of safety.
After completing the CAPTCHA, the victim is directed to a page that spoofs the interface of Zoom, another ubiquitous teleconferencing tool. The page informs the user that they must register to attend the "private" event and prompts them for their work email address. To proceed, users are given the option to "Continue with Google" or "Continue with Microsoft." Selecting either of these triggers a second CAPTCHA page.
The researchers noted that these dual CAPTCHA layers are highly effective at detecting and blocking analysis from VPN services, Tor browsers, and other security scanners. "A CAPTCHA was built to keep bots out," Dickson pointed out. "Here, it keeps the security scanners out and lets the victims in."
The final stage of the attack involves the theft of credentials through spoofed login pages. While the Google and Microsoft spoofs are both "remarkably similar" to their legitimate counterparts, they behave differently. The fake Microsoft page is designed to collect both usernames and passwords, using a deceptive URL—office[.]com-access-sharepoint[.]com—that mimics real Office and SharePoint domains. The fake Google page, using the domain goo[.]gle-access-eventregister[.]icu, currently only captures usernames. The researchers suspect this may be due to a misconfiguration on the attackers’ part, rather than a lack of intent.
Once the user submits their information, the site provides no confirmation of registration and does not redirect to a legitimate vendor page. However, by that point, the damage is done. The attackers have successfully exfiltrated the credentials, providing them with a foothold to compromise accounts, access sensitive corporate documents, or launch further internal attacks within the enterprise.
It’s not the employees’ fault
The success of such campaigns highlights a systemic vulnerability in how modern business is conducted. Frank Dickson noted that the AI industry’s own habits—such as the frequent use of gated previews, "early access" lists, and waitlists—actually train employees to expect exactly this type of exclusive, invitation-only communication. When the legitimate industry standard involves jumping through hoops for access, a phishing scam that mirrors that process becomes much harder to identify.
Dickson advised organizations to look for the "cracks" left by the attackers. He pointed out the jarring inconsistency of the "brand salad" presented to the user: the email claims to be from "nthropic" but uses GoTo Webinar branding, the registration page looks like Zoom, and the final login is via Google or Microsoft. "That’s three brands in three clicks," he noted, a red flag that suggests the process is not a unified corporate event.
Furthermore, the malicious domains are specifically designed to be "skimmed, not read." In the heat of a busy workday, an employee is unlikely to carefully parse a URL like office[.]com-access-sharepoint[.]com, which contains enough familiar keywords to pass a cursory glance.
While the current version of the Google spoof may contain a configuration error regarding password collection, Dickson warned that the threat actors behind the campaign will likely fix such issues in future iterations. The sophistication of the mimicry is only expected to increase as attackers themselves begin to use AI tools to refine their scripts and design.
The researchers at Cofense suggested several practical steps for users, such as checking for consistent branding across the entire journey from email to login and verifying that the URL and domain match the service being used. However, Dickson emphasized that the burden of defense should not rest solely on the individual worker.
"Training helps, but I would not hang the defense on an employee squinting at a URL at 4:45 on a Friday," Dickson said. He argued that the value of a stolen password is a leadership decision, not a personal failure of the employee. If a single stolen password can lead to a catastrophic breach, the organization’s security architecture is fundamentally at fault.
To mitigate these risks, Dickson and the researchers advocate for the implementation of phishing-resistant multi-factor authentication (MFA), such as passkeys. Unlike traditional SMS-based or app-based MFA, which can still be bypassed through sophisticated proxy attacks, passkeys and hardware security keys make a stolen password virtually useless to an attacker.
Ultimately, this campaign serves as a stark reminder of how easily trusted digital environments can be turned against users. By using legitimate-looking tools and exploiting the current cultural obsession with AI, attackers have found a way to bypass both technical filters and human skepticism.
"Bottom line: This campaign uses trusted tools to defeat defenses built on trust," Dickson concluded. "The employee should not be the last line of defense." As the AI boom continues to dominate corporate interest, organizations must ensure that their security infrastructure evolves as quickly as the lures used to target their staff.
