In the high-stakes world of enterprise artificial intelligence, "FOMO"—the fear of missing out—has become a potent psychological weapon. A newly identified, highly sophisticated phishing campaign is currently exploiting the industry’s culture of gated previews and exclusive access, targeting ambitious employees with fraudulent invitations to private Anthropic AI demonstrations.
By weaponizing the branding of reputable platforms like GoTo Webinar and Zoom, and utilizing double-layered CAPTCHA defenses to bypass automated security scanners, attackers are successfully harvesting Google and Microsoft corporate credentials. This campaign represents a maturation of social engineering, where the allure of cutting-edge technology is used to lower the defensive guard of even the most diligent employees.
The Mechanics of the Deception: How the Campaign Unfolds
The campaign is characterized by its high degree of visual polish and its cynical exploitation of corporate communication norms. According to researchers at Cofense, the attackers have carefully crafted a multi-stage funnel designed to appear entirely legitimate to the average user.
The Initial Lure
The attack begins with an email masquerading as a high-priority invitation to an "exclusive, invitation-only virtual session." The content promises a "focused conversation on AI systems, research, and the future of frontier models." By citing Anthropic—a company currently at the center of the AI news cycle—the attackers tap into the professional aspirations of employees who want to stay ahead of the curve.
The Branding Mismatch
The email arrives with the professional sheen of a GoTo Webinar notification. However, beneath the surface, the cracks begin to show. The sender address, customercare@gotowebinar[.]com, attempts to mimic the platform, but the sender name is displayed as "Austin from nthropic"—a minor but telling typo where the leading "A" is missing. This type of subtle error is a classic hallmark of mass-produced phishing, yet it is easily missed by an employee scanning their inbox during a busy workday.
The "Double-Gate" Defense Evasion
Perhaps the most innovative aspect of this campaign is its use of double-layered CAPTCHA challenges. Once a user clicks the registration link, they are not immediately taken to a phishing site. Instead, they encounter a Cloudflare CAPTCHA.
This serves two purposes:
- Psychological Validation: Because CAPTCHAs are synonymous with security and legitimacy, the user assumes they are entering a protected environment.
- Security Evasion: The CAPTCHA effectively blocks automated security crawlers, sandboxes, and VPN-based scanners from analyzing the site’s content. By the time a user arrives at the fake Zoom-branded registration page, the malicious infrastructure has effectively been "hidden" from the organization’s automated perimeter defenses.
Chronology of the Attack Path
To understand the scale and persistence of this threat, it is necessary to view the attack as a linear progression:
- Phase 1 (The Hook): The target receives an email formatted as a GoTo Webinar invite, leveraging the reputation of Anthropic.
- Phase 2 (The Redirect): The user clicks a link, which has been obfuscated through URL shorteners and subdomains to bypass email filters.
- Phase 3 (The Scanner Bypass): The user is presented with an initial CAPTCHA. This prevents security tools from verifying the destination URL.
- Phase 4 (The Spoof): A convincing Zoom-style registration page appears, requesting the user’s work email.
- Phase 5 (The Credential Harvest): The user clicks "Continue with Google" or "Continue with Microsoft." A second CAPTCHA appears, further filtering out security bots.
- Phase 6 (The Exfiltration): The user is directed to a pixel-perfect clone of a Google or Microsoft login portal. Credentials are submitted, and the user is left on a dead-end page, with no confirmation or redirect, while the attackers capture the login data.
Supporting Data: The Anatomy of a Spoof
The technical infrastructure supporting this campaign demonstrates a deliberate, albeit occasionally flawed, strategy.
- Google Spoofing: In some instances, the fake Google portal only captures usernames. Security researchers believe this is likely due to a misconfiguration in the attacker’s toolkit, but they warn that this is a temporary state; the threat actors are likely to iterate and perfect this component quickly. The domain,
goo[.]gle-access-eventregister[.]icu, mimics the structure of legitimate Google services while using a suspicious top-level domain. - Microsoft Spoofing: The Microsoft-targeted pages are more comprehensive, capturing both usernames and passwords. The URL,
office[.]com-access-sharepoint[.]com, is designed to be skimmed rather than read, preying on the human tendency to look only at the first few characters of a web address.
Frank Dickson, a lead researcher at Dickson Research, emphasizes that this campaign is a "triple-threat" of identity theft: "The attackers are leveraging three major brands—GoTo Webinar, Zoom, and Google/Microsoft—all within the span of three clicks. It is a masterful display of how modern phishing relies on the user’s trust in established, familiar ecosystems."
Implications for Enterprise Security
The success of this campaign carries significant implications for how corporations view their internal defensive strategies.
The Fallacy of the "Human Firewall"
For years, organizations have relied on security awareness training to turn employees into the "last line of defense." However, this campaign proves that even a tech-savvy, well-intentioned employee can be deceived when the lures are sufficiently personalized and the technical execution is high-fidelity.
"We are training employees to expect these kinds of invitations," says Dickson. "By creating a culture of exclusive, gated AI previews, the industry has inadvertently provided the perfect cover for these actors. It is not the employee’s fault; it is a structural failure in how we authorize and verify the communications we send and receive."
The Move Toward Phishing-Resistant MFA
The primary takeaway for CISOs and IT departments is the inadequacy of traditional, knowledge-based credentials. Passwords, regardless of complexity, are inherently harvestable. The industry is reaching a consensus that organizations must move toward FIDO2-compliant, phishing-resistant multi-factor authentication (MFA), such as hardware security keys or passkeys.
Unlike standard SMS or push-based MFA, which can be bypassed via prompt bombing or proxy-based phishing, phishing-resistant MFA cryptographically ties the authentication process to the specific domain. Even if an employee enters their credentials into a perfect clone of a Microsoft login page, the hardware key will refuse to sign the request because the domain does not match the legitimate source.
Moving Beyond the "Last Line of Defense"
The conclusion drawn by cybersecurity experts is clear: the employee should never be the primary defense against a sophisticated campaign of this nature. The burden of security must shift back to the technical infrastructure and policy.
- Adopt Hardened MFA: Phishing-resistant MFA is the single most effective tool against credential harvesting. If the authentication process cannot be proxied, the attack chain is broken.
- Domain Reputation Filtering: Organizations must implement more aggressive filtering of suspicious TLDs (like
.icuor.top) and utilize advanced email security gateways that can perform deeper inspection of URLs that utilize CAPTCHAs. - Internal Verification Protocols: Companies should establish clear internal channels for vetting "exclusive" events. If an employee receives an invite to a private AI preview, there should be a centralized process to verify the sender’s legitimacy before a single click is made.
- Behavioral Analysis: Security teams should monitor for anomalous login patterns, such as a user suddenly attempting to access sensitive documents from an unrecognized device shortly after an unusual login event.
As the AI industry continues to expand, the "Fear of Missing Out" will only become more acute. Attackers are well-aware of this, and they are already preparing the next iteration of this campaign. Whether the lure is a private AI preview today or a breakthrough in quantum computing tomorrow, the methodology remains the same: exploit the trust we place in the tools we use every day.
For the modern enterprise, the path forward is not found in scolding employees for their curiosity, but in building systems that render stolen passwords useless. The days of assuming that a breach is someone else’s problem have long since passed; in the era of AI-driven social engineering, the only safe assumption is that the next "exclusive invitation" in your inbox might be the most dangerous one you ever receive.
