How AI is driving email phishing and how to beat the threat [Q&A]

Among all of the various forms of cyberattack phishing attempts delivered by email are still one of the most common.

What’s more AI is making these attacks more effective, because you can no longer rely on looking out for dodgy grammar or other signs that a message may not be what it seems.

We spoke to Alan Lefort, co-founder and CEO of StrongestLayer, to find out how AI is improving phishing attempts and what organizations can do to combat the threat.

BN: What security vulnerabilities are enterprises encountering as AI-generated phishing attacks scale beyond what traditional email security systems were designed to handle?

AL: The rise of generative AI has permanently altered the cyber threat landscape. Attackers now use jailbroken LLMs to craft thousands of personalized, context-aware phishing emails -- at scale and in seconds. These messages evade traditional detection because they don't reuse content, follow predictable structures, or exhibit statistical anomalies.

Traditional email attacks were volume-based and obvious. AI changes the game completely -- attackers can now generate personalized emails that understand organizational hierarchies, mimic communication styles, and reference actual business processes. They're not just sending generic phishing anymore; they're crafting attacks that feel authentic to specific organizations.

The biggest vulnerability is that traditional email security has always been about detecting ‘bad’ technical patterns, but AI attacks look technically good while being maliciously intentioned. Legacy systems operate like a prosecutor-only court system -- they can only hunt for guilt with no mechanism to prove innocence, creating the classic false positive/false negative tension that can never be solved within a prosecutor-only architecture.

For example, we recently detected an AI-crafted HR impersonation campaign that used Unicode bidirectional text manipulation to display ‘Official Notification: Performance Evaluation Access’ while the actual text was reversed, completely evading regex pattern matching. The attack combined typo-squatted domains, psychological manipulation about ‘terminations marked in red,’ and perfect mimicry of internal HR communications. Traditional security systems will attempt to detect suspicious Unicode usage, but this creates massive false positives for legitimate Arabic, Hebrew, and other right-to-left language business communications -- forcing organizations to choose between missing sophisticated attacks or constantly investigating legitimate international correspondence.

BN: As threat actors increasingly leverage generative AI to create personalized, context-aware phishing campaigns, how is this changing the detection challenges that mid-market organizations face?

AL: Mid-market organizations are caught in a growing security gap. They face enterprise-level attack sophistication without enterprise-level resources. With limited SOC capacity, no in-house threat intelligence, and growing alert fatigue, these teams are left exposed to AI-enabled Business Email Compromise (BEC), vendor fraud, and deepfakes that legacy security tools miss entirely.

Organizations with complex vendor relationships and approval processes are at highest risk -- particularly financial services, healthcare, and legal firms. Mid-market companies face a specific challenge: they have enterprise-level obligations and complexity but without the security staffing or budgets to match. They're sophisticated enough to be attractive targets but resource-constrained enough that advanced attacks succeed.

Email is where trust decisions happen daily. Employees regularly receive requests from vendors, partners, and colleagues that require judgment calls about legitimacy. AI-powered attacks exploit exactly these trust relationships by appearing to come from known entities with plausible requests.

BN: Rules-based email security has been the standard for decades, but you've said ‘rules failed’ in the age of AI threats. How does the shift from pattern-matching to intent-based detection alter the security posture of email environments?

AL: The traditional ‘Patient Zero’ model -- detect, analyze, and react -- has collapsed under the weight of modern, polymorphic threats. These attacks are generated on the fly and optimized in real time. The only viable defense is a system that understands malicious intent, not just matches known indicators.

We built the first LLM-as-master architecture, fundamentally different from vendors who bolt LLM features onto existing pattern-matching systems. Our breakthrough is dual evidence collection that breaks the prosecutor-only paradigm entirely by having every email get its day in court. Our system acts as both public defender and prosecutor, while an impartial LLM judge weighs evidence and renders verdict.

This shift fundamentally changes email security economics. Instead of legacy systems that see urgent language, large amounts, after-hours timestamps and quarantine legitimate business communications, our dual evidence architecture runs parallel investigations that focus on stable indicators that persist regardless of attack novelty -- business legitimacy patterns and malicious intent patterns remain consistent even when attack methods are completely novel.

BN: With business email compromise attacks becoming more sophisticated through AI generation, what practical considerations should security teams evaluate when comparing traditional signature-based versus reasoning-driven email protection models?

AL: Traditional signature-based systems deliver five to 30 percent precision compared to AI-powered reasoning systems that achieve 95 percent precision with dramatically reduced false positives. Legacy email gateways are mostly about after-the-fact detection, while reasoning-driven systems provide real-time threat detection and risk prioritization.

The key practical consideration is that while AI makes attack generation infinitely scalable, it doesn't solve the business logic problem for attackers. AI can craft perfect grammar, mimic communication styles, and reference public information about organizations. But it can't perfectly understand internal business processes, approval workflows, contract cycles, and relationship dynamics across every organization simultaneously.

Security teams should evaluate systems that focus on business reasonableness rather than technical patterns. As AI attacks get more sophisticated in their technical presentation, they often become more desperate in their business logic -- creating urgency where none should exist, bypassing normal processes, or requesting actions that don't align with established relationships.
Unlike legacy products that often take days to configure and tune, reasoning-driven systems require no rules or thresholds and adapt instantly to your environment, bringing the judgment of a threat analyst to every inbox at machine speed.

BN: How might the convergence of AI-powered attack automation and the growing sophistication gap between enterprise and mid-market security resources shape the evolution of email threat detection over the next two to three years?

AL: By 2026-2027, sophisticated AI-enhanced attacks will become mainstream rather than advanced persistent threat territory. The tools and techniques will become commoditized, dramatically expanding the threat actor pool. We'll also see attacks expanding beyond email to all business communication channels -- Teams, Slack, mobile messaging, with attackers orchestrating campaigns across multiple channels.

This evolution will make tools like reasoning-driven email protection a default layer of control for enterprise environments. Major platforms like Microsoft Entra Internet Access are already rolling out inline, pre-submission controls for AI traffic. When you pair that with industry consensus around AI-generated threats, it's reasonable to expect AI-aware detection around communications to become a default control layer alongside identity and access management and endpoint detection and response.

The other major shift will be attacks that exploit AI adoption within organizations. As companies deploy AI tools for business processes, attackers will target those integrations and data flows. The ultimate evolution is reaching a point where organizations don't think about email security as a separate problem -- it's just built into how business communication works, protecting them automatically while enabling productivity in an AI-powered world.

Image credit: Josepalbert13/Dreamstime.com