The lure and the mechanism
According to security firm Cofense, a phishing campaign is underway that uses Anthropic's name to steal corporate credentials. Victims receive an invitation to a supposed private preview, reserved for a select few, of the company's latest technologies; the real target would be Google and Microsoft passwords.
The message mimics a webinar invitation and exploits the GoTo Webinar brand, a platform whose convocation emails are normally not blocked by security systems. Those who click to register are directed to a compromised address; from there a multi-stage path unfolds: a Zoom-style registration page, CAPTCHA checks designed to keep out scanners, VPNs, and Tor-based analysis, and finally fake Google and Microsoft login screens.
What is harvested and the clues
According to the cited analysis, Microsoft users give up both username and password, while the fake Google page appears to capture only the username, probably due to a configuration error. In any case, a stolen login would allow attackers to move within corporate systems.
Some recognizable clues remain: a sender signed "Austin from nthropic," logos of different services mixed on the same page, and deceptive domains such as office.com-access-sharepoint.com. These are signals described by the source, not a guarantee of detection for users.
Why Anthropic and how to defend against it
Frank Dickson of Dickson Research explains that the invitation sells exclusivity and leverages the fear of missing out: every organization is trying to figure out where to place artificial intelligence, and a private preview of Anthropic's latest developments is the invitation an ambitious employee would want to accept. According to Dickson, attackers did not choose Anthropic at random: they take the hottest name of the moment, and next month it could be another.
Anthropic had already been targeted in other operations. In October, Proofpoint attributed to the China-aligned group TA419 campaigns against AI policy experts in the United States, with emails impersonating an Anthropic executive and a former White House official. That case was targeted and espionage-type; the one described by Cofense instead targets a broad corporate audience.
The defenses indicated by experts go beyond individual vigilance. Dickson recommends phishing-resistant multi-factor authentication, credential resets, and independent verification of every AI-related invitation, without relying on employees' ability to recognize every fake. Anyone who entered data on the fake page should revoke active sessions, check for suspicious access, and change their password. The underlying picture remains: every market-anticipated announcement, from a new model to a waiting list, becomes an opportunity for scammers, and companies that treat AI-related invitations as a risk category, with official verification channels, reduce the room for maneuver. The indications come from the source and the experts cited; autonomous data on the number of victims or the duration of the campaign are not available in the material, elements that remain to be verified.
