Attackers use AI-generated fake login pages to steal login codes and hijack sessions

Attackers can use generative AI to build fake login pages that imitate trusted brands, capture multifactor authentication codes, and hijack authenticated sessions, The Hacker News reported Oct. 27, 2026. Fraudulent login pages can also closely copy company logos, layouts, and login prompts from sites employees use regularly. Taking an active session token lets attackers seize an account without a password or another multifactor authentication challenge, even with multifactor authentication turned on.

A white hat hacker based in Budapest demonstrated that ChatGPT could create a fake employee login page within two minutes using a single command, and the logo on the page closely resembled Microsoft’s actual branding. The demonstration appeared as part of a TV Asahi segment, showing how quickly generative AI can recreate familiar authentication portals that employees use daily.

The Hacker News reported Oct. 27, 2026, that criminals customize these messages for particular workers or companies, using personalized IT alerts that send staff to fake login pages carrying company branding. Scammers can generate customized e-mail and SMS messages in bulk by feeding scraped personal data into large language models, which strips out common spelling mistakes and template language so each message is distinct. It might reference an employer, a recent LinkedIn post, or a mutual acquaintance, and scraped data from data brokers can make the appeals more convincing by referencing real relationships.

These pages may show a person’s name, account details, or past interactions to make the page feel personal and trustworthy. AI-generated text, images, and videos help scammers build login pages that look and behave like genuine sites, which simplifies password theft.

Lookalike domains often carry subtle spelling changes or extra characters and appear next to authentic login pages with focused messaging, making phishing harder to spot. Attackers depend on web addresses that closely resemble legitimate ones. A secure HTTPS connection does not by itself prove a website is genuine.

Some fake login pages capture multifactor authentication codes through replicated multi-step authentication and use the codes before they expire, intercepting the one-time password at the moment the victim enters it. Others hijack authenticated sessions, which lets attackers access corporate accounts even with additional security measures in place because the attacker rides the victim’s authenticated session rather than needing the password. Criminals who seize a live session token can take over an account without needing a password or another multifactor authentication prompt, even if multifactor authentication is turned on, because they seize session cookies or authentication tokens once the person has logged in and then reuse those tokens elsewhere.

In an Adversary-in-the-Middle setup, the attacker places a fake site between the user and the genuine service and relays authentication requests in real time. After the person completes login, the attacker can grab a session cookie or token and access the account without another MFA challenge. This session hijacking works even when MFA is enabled because the attacker intercepts the authenticated session rather than the password itself, and the legitimate service sees a valid token from what appears to be the user’s device. Entering a code into a fake page does not protect the account if the attacker can capture the resulting session.

Human operators add another layer of flexibility to the attack. On pages tied to fake AI services, an operator could respond to victim input by switching authentication screens and requesting extra MFA codes.

A news report from The Register dated Oct. 7, 2026 described fake AI services that drew corporate ad managers into Browser-in-the-Browser pages, which paint a fake browser window inside a web page and mimic the browser’s own interface. The fraudulent services carried names including ChatGPT, Claude, Gemini, Perplexity, and Meta Muse. Meta Muse was announced by Meta in September 2026 and fake ad management sites using it appeared quickly.

The pages showed a genuine Google address in a fake address bar even though the content came from an attacker-controlled site. Researchers observed one system collecting input from roughly 200 different e-mail addresses across about one month, which proves data was entered but does not prove accounts were compromised.

On Mar. 4, 2026, Microsoft Threat Intelligence published research on the Tycoon2FA phishing service. The service copied the look of Microsoft 365 and Gmail sign-in screens and went after passwords and session details at the moment of authentication, intercepting credentials as users entered them into what appeared to be familiar corporate and consumer portals. Microsoft stated that it worked with police and security agencies across several countries to take down the Tycoon2FA infrastructure, acting against the underlying hosting and delivery network that supported the phishing pages.

On Sept. 9, 2026, Microsoft Security Research released an analysis focused on social engineering aimed at passkey users. In the report, attackers pose as IT staff and send users to Microsoft login screens that look nearly identical to the real ones, exploiting trust in internal support communications to bypass suspicion. Sometimes the attackers add their own authentication method to the victim’s account, which means resetting the password is not enough to end the compromise because stolen sessions can survive a password change.

According to the 2026 Verizon Data Breach Investigations Report, phishing accounted for 44 percent of the AI-assisted initial-access techniques examined. Note reported that a USENIX Security 2026 study involving 7,700 people measured link click rates for AI-personalized phishing e-mails at roughly three times the rate for generic e-mails. The figure reflects experimental click behavior rather than confirmed credential theft or financial harm, yet it indicates that personalization sharply increases engagement. Security.org explains that AI identity theft relies on tools such as generative AI and machine learning, and that the goals range from stealing and fabricating identities to exploiting them for financial fraud or account takeover.

Models trained on leaked credential data can guess probable passwords. AI-driven CAPTCHA solving and browser automation make large-scale credential testing difficult to tell apart from normal login traffic, which allows automated attacks to blend in with normal user behavior. Security.org also notes that these tools can fabricate synthetic identities that do not match real people, letting attackers fabricate supporting evidence for fake accounts. Brett Cruz, a cybersecurity expert, said, “Account takeover is table stakes for serious identity fraud.”

Security teams saw waves of automated credential-stuffing attempts against Facebook, Instagram, and LinkedIn accounts spanning late 2025 and early 2026, using breached username and password pairs fed through automated bots. LinkedIn issued a warning to its 1.2 billion users about an increase in policy violation attacks intended to manipulate reporting systems and force changes to content or verification.

Sift data cited by NHI Management Group shows that 83 percent of organizations saw at least one account takeover in 2025. The group projects account takeover losses will reach 17 billion dollars, up from 13 billion dollars the prior year. The Sift Digital Trust Index for Q3 2025 showed a 4 percent year-over-year rise in account takeover attempts, with the sharpest jumps seen in e-commerce, fintech, and internet-related software. NHI Management Group says identity, fraud, and IAM teams should treat account takeover as an ongoing trust issue spanning login, session and transaction stages. That means account takeover is more than a one-time authentication event.

Digital Transactions reported Oct. 26, 2026, citing Microblink data, that 30 percent of fraud failures involved fake U.S. identity documents. Microblink, a fraud-prevention firm based in Brooklyn, N.Y., says false documents can help criminals enter trusted accounts. Microblink argues that older identity models assume trust remains fixed, whereas modern fraud campaigns travel across the customer journey and strike when trust is presumed rather than verified.

Not every fake-login or session-hijacking attack is executed by generative AI. Some attacks rely on existing phishing services and human operators instead, and building a fake page does not, by itself, demonstrate that credentials were stolen or systems were infiltrated. A fake page alone does not prove credential theft or system infiltration.

Passkeys and FIDO2 security keys tie cryptographic credentials to legitimate website domains, which reduces the chance of credential theft on fake login pages since the credentials fail on a lookalike domain that does not match the registered origin. CISA recommends FIDO-based authentication as a phishing-resistant method. Security teams need controls that evaluate suspicious websites and catch credential-harvesting attempts before users enter login information, because traditional defenses have gaps even when multifactor authentication is present. These controls can intercept the phishing page before the user ever types a code or password.