The threat landscape for credential-based attacks has fundamentally shifted. AI does not introduce entirely new attack categories, but it dramatically compresses the time and cost between information gathering and successful exploitation — enabling threat actors to scale highly targeted, personalized attacks that previously required significant manual effort per victim. The result is that traditional authentication factors that security teams have relied on for years are no longer sufficient on their own.
The Credential Problem Is Getting Worse
Stolen credentials are involved in 44.7% of breaches, according to the Verizon Data Breach Investigations Report. That figure has been stubbornly persistent across years of security investment in MFA, phishing-resistant authentication, and security awareness training. The reason is straightforward: the defenses evolved, and so did the attacks.
AI-enhanced phishing campaigns can now generate thousands of highly personalized, contextually convincing messages with minimal effort. By pulling publicly available data — LinkedIn profiles, company press releases, industry event attendee lists — threat actors automatically build detailed victim profiles and tailor lures to specific roles. A finance team member receives a spoofed supplier invoice referencing a real vendor relationship. An IT administrator receives a cloud access alert matching their actual service stack. The generic, grammatically awkward phishing emails that awareness training historically taught users to spot are increasingly a thing of the past.
MFA Is Being Bypassed Routinely
Multi-factor authentication, long the primary defense against credential compromise, now faces several well-documented bypass techniques that AI-enhanced tooling makes more accessible at scale:
Adversary-in-the-Middle (AiTM) Proxies sit transparently between the victim and the legitimate service. When a user enters their credentials and MFA code on the phishing page, the proxy relays them to the real service in real time, completing authentication and capturing the resulting session cookie. The attacker never needs the password or OTP again — they have a valid authenticated session.
Push notification fatigue and abuse exploits the friction of MFA approval by sending repeated authentication push requests until a user approves one, or by using social engineering to request approval ("I'm locked out, can you approve the MFA prompt?").
Session cookie theft is the most technically elegant bypass: rather than defeating MFA at authentication time, attackers target the session tokens issued after a successful MFA authentication. Stealing a valid session cookie sidesteps MFA entirely for the remainder of the session's lifetime. Infostealer malware families specifically target browser session storage for this purpose.
A real-world example illustrates the risk: in one documented incident, attackers exploited Restream.io credentials that had been sitting in infostealer dumps for approximately one month. The stolen credentials bypassed MFA because the associated session cookies remained valid — the organization had no signal that a threat actor was operating in their environment until observable damage occurred.
IP Geolocation and Network Trust Are Defeated
Traditional threat signals like IP geolocation and network reputation have become nearly useless as primary authentication trust signals. Residential proxy networks position exit nodes geographically near victims, making attacker traffic appear local. Mobile networks share IP addresses across thousands of users. Compromised endpoints serve as unwitting relays. An attacker operating from Eastern Europe can appear to be authenticating from the same city as their target.
NIST SP 800-207, the Zero Trust Architecture guidance, is explicit: organizations "should not grant implicit trust based solely on physical or network location." If you are still using IP-based allow lists or geolocation as a primary trust signal, those controls are providing false confidence.
The Case for Device Trust
The next essential authentication layer is device trust — requiring not just verification of who you are (credentials and MFA), but what device you are using and whether that device meets current security health requirements. The logic is straightforward: while credentials and session tokens can be stolen and replayed from any device, hardware-bound device attestation cannot.
An effective device trust framework has four components:
Hardware-anchored device registration ties authentication to the specific device using hardware security modules (TPM chips, Secure Enclave). A credential combined with hardware attestation from a registered device is fundamentally harder to steal and replay than a credential alone.
Continuous session-level evaluation monitors both user identity and device health posture throughout a session, not just at login. A device that passes health checks at 9:00 AM but is compromised by malware at 10:30 AM should trigger re-evaluation — not continue operating under the trust established at initial authentication.
Risk-proportionate enforcement applies graduated responses rather than binary block/allow decisions. Low-risk actions (reading email, viewing dashboards) might proceed with standard auth; high-risk actions (wire transfers, admin configuration changes, bulk data exports) trigger step-up verification or require a health-verified device.
User-guided self-remediation provides clear paths for users to bring non-compliant devices back into compliance — updating endpoint protection, applying patches, reconnecting to device management — rather than simply blocking access with no explanation or path forward.
Practical Starting Points
Organizations beginning to implement device trust can start incrementally:
- Audit your current MFA implementation. Push-based MFA (authenticator app approvals) is significantly weaker than phishing-resistant options (FIDO2 passkeys, hardware security keys). Migrate high-value accounts first.
- Enable conditional access policies that factor in device compliance state. Microsoft Entra ID, Google Workspace, and Okta all support device-based conditional access.
- Implement session lifetime controls. Enforce re-authentication after session inactivity periods and at defined maximum session durations to limit the value of stolen session tokens.
- Scan Active Directory for compromised credentials. Tools like Specops Password Auditor can identify accounts with passwords appearing in known breach databases — these accounts should be treated as already compromised.
- Deploy endpoint detection that flags infostealer activity. Session cookie theft happens on the victim's device before the cookie is ever used. Catching infostealer execution is the earliest possible point of intervention.
The fundamental shift required is moving from a model that asks "do you know the right secrets?" to one that asks "are you who you claim to be, on a device that meets our security requirements, behaving in a way consistent with your normal patterns?" Credentials answer the first question. Device trust, behavioral signals, and continuous evaluation are what it takes to answer the second.