SCOTT CHILLE / CYBER PULSE / FIELD GUIDE
CYBER PULSE · WEEKLY THREAT INTELLIGENCE · SCOTT CHILLE
Identity & The Human Element

They Didn't Hack In. They Logged In.

AI hasn't just made phishing more convincing. It's made every step of stealing a credential, and everything an attacker does with it afterward, faster, cheaper, and harder to catch.

82.6%
of phishing emails are now AI-generated, up from a small fraction just a few years ago
4.5x
higher rate of credential entry once a target clicks an AI-generated phishing message, versus a traditional one

The new front door is a phone call

Email got harder to fake convincingly for exactly one reason: filters got good at catching it. Voice didn't get the same defense.
442%
surge in voice phishing (vishing) from the first half to the second half of 2024, the fastest-growing attack vector tracked
3 sec
of audio is enough to clone a voice convincingly with current AI tools, roughly the length of a voicemail greeting
Case in point

The helpdesk is the target now, not the employee

Groups like Scattered Spider have shifted their playbook away from technical exploits toward calling helpdesks directly, impersonating an employee locked out of their account, and talking a support agent through an MFA reset. It worked against MGM Resorts in 2023 at a cost north of $100 million, and by 2026 the same playbook has moved decisively into financial services and insurance. The attacker doesn't need to beat your MFA. They need to convince one person to reset it.

Once they're in, MFA doesn't save you

The credential was never really the target. Access is.
01

Adversary-in-the-middle kit intercepts the login

Off-the-shelf phishing kits proxy a real login page in real time, capturing the password and the session token behind it.

AiTM
02

Session token theft bypasses MFA entirely

The attacker doesn't need the second factor. They already have the authenticated session.

80% of MFA-bypass breaches
03

Stolen credential seeds a much larger compromise

A single harvested login becomes lateral movement, privilege escalation, or a ransomware foothold.

Legitimate access, no alert
The Detection Gap

You already know this is a problem. Testing for it is a different story.

Only 40% of CISOs name credentials as the data type that would cause the most damage if stolen, ranked no higher than financials or trade secrets. Yet once credentials are abused, perimeter controls and patching offer almost nothing, the attacker is already operating with legitimate access.

Frequently test AD credential attacks
38%
Fully confident detecting directory attacks
<50%

It isn't just human identity anymore

The same root failure, identity without verification, is now showing up on the agentic AI side too.

Human flank

A helpdesk agent trusts a confident voice on the phone and resets an MFA device that was never actually lost. Trust exploited through fluency and process, not a technical flaw.

Agent flank

An AI agent inherits a cached credential or a delegated token with no distinct identity of its own. Per OWASP's Agentic Top 10, this is one of the most common paths attackers use to escalate access inside agentic systems.

What actually closes the gap

01

Deploy phishing-resistant MFA on every high-value account

Hardware keys or passkeys remove the credential-theft surface entirely. Password and SMS-based factors are the weakest link left standing.

02

Treat the helpdesk as a privileged access point, not a courtesy desk

Require a phishing-resistant identity check before any MFA reset, device change, or account recovery action, with no agent discretion to bypass it under pressure.

03

Bind sessions to devices

Token binding makes a stolen session unusable from anywhere but the original device, closing the AiTM path even after credentials are phished.

04

Measure click-to-report, not click rate

The median time to click a phishing link is seconds. The median time to report one is closer to half an hour. That gap is where damage happens.

05

Extend awareness training beyond email

If your simulation program only covers email, it has a blind spot the exact size of a phone call.

The password isn't the weak link. The verification process is.