Zero Trust in a SaaS and AI World
The Browser Is the Enterprise Perimeter
For years, we said the perimeter was dead. We moved to cloud services, adopted SaaS, embraced remote work, and redesigned identity around Zero Trust. Yet if you look at the way many recent compromises unfold, a different perimeter becomes visible: modern attacks often begin not with network exploitation, but inside the browser.
An engineer pastes proprietary code into a generative AI tool, a finance user approves a malicious OAuth application, or an attacker replays a stolen SaaS session token and bypasses the MFA event entirely. These are no longer unusual edge cases. They are recognisable incident patterns in environments where most daily work happens through web applications.
The browser is therefore no longer just a rendering engine. It has become the operating environment for SaaS, identity, and AI-driven workflows, yet many organisations still treat it as a trusted surface rather than a security boundary in its own right.
The Threat Shift
Identity is widely described as the new perimeter. That position is formalised in NIST SP 800-207 Zero Trust Architecture, which defines continuous verification and policy enforcement close to the resource as foundational principles.
But SaaS changed something important. Data now lives inside browser sessions, OAuth grants can provide API access without exposing a password, and session tokens can carry authenticated access beyond the original MFA event. Generative AI tools add another route by accepting unstructured prompts that may contain intellectual property or regulated data.
Industry reporting reflects this shift. The Microsoft Digital Defense Report highlights adversary-in-the-middle phishing and token theft. Incident analysis in the Mandiant M-Trends Report and the CrowdStrike Global Threat Report shows attackers increasingly leveraging valid session tokens to persist inside cloud and SaaS environments without triggering traditional alerts.
When attackers compromise a SaaS session, they may not need lateral movement in the traditional network sense because they already have authenticated access to the application layer. Session hijacking, OAuth abuse, and shadow AI usage are all browser-native paths that sit awkwardly outside security models built primarily around network boundaries and endpoint processes.
Mapping the Pattern
If we map these behaviours to the MITRE ATT&CK Enterprise Matrix, they align with:
- Valid Accounts
- Abuse of Delegated Credentials
- Exfiltration Over Web Services
The tactics are not new, but the execution layer has changed. If a Zero Trust strategy stops at device compliance and MFA, it may harden the front door while leaving the interaction layer largely implicit. The browser has become an operational perimeter whether we deliberately planned for it or not.
In Part 2, we will look at how to enforce Zero Trust at the browser layer using deterministic controls that reduce risk without making ordinary work unnecessarily difficult.
👉 Read Part 2: Enforcing Zero Trust at the Browser Layer