Attackers Prioritized Optimization Over Innovation With AI in 2025

Security researchers observed that threat actors leveraged AI to scale traditional attack methods rather than develop novel techniques, emphasizing the need to strengthen fundamental defenses.

While headlines touted AI-powered cyber threats throughout 2025, attackers demonstrated remarkable pragmatism: They systematically optimized existing attack vectors rather than pursuing radically new methods. According to OX Security Research Team Lead Moshe Siman Tov Bustan, this revealed a critical lesson for defenders.

"Attackers didn't reinvent their playbook when AI arrived—they automated it," Bustan explains. "They're still exploiting supply chains, phishing developers, and bypassing app store reviews. The difference is they're executing these attacks with one-tenth the resources."

Three Attack Vectors Where AI Amplified Existing Tactics

Supply Chain Compromise at Scale: The Shai Hulud NPM campaign exemplified how AI enables smaller threat actors to create cascading impacts. What previously required organized crime groups can now be executed by individuals. Attackers publish legitimate packages to build trust over months or years, then trigger malicious payloads across entire dependency trees with minimal effort. This pattern mirrors the earlier XZ Utils compromise, demonstrating attackers' willingness to play the long game.
Hyper-Targeted Phishing Operations: Human vulnerability remains the most reliable attack surface. Recent npm incidents showed how a single developer's compromised credentials—obtained through phishing—can poison packages with tens of millions of weekly downloads. AI enables attackers to craft more convincing lures and automate credential harvesting at unprecedented speed, turning individual mistakes into widespread incidents before mitigations deploy.
Malware in Official Repositories: Malicious Chrome extensions stealing ChatGPT and DeepSeek conversations revealed persistent gaps in automated store reviews. Despite available solutions—like the granular permission models used in Android and iOS—Chrome extensions still operate on binary "all-or-nothing" access requests. As Bustan notes: "If an extension requests permission to read all website data, it's either malicious now or will be after an update."

Defensive Priorities for 2026

Security teams should refocus on foundational controls:

Implement Granular Permissions: Adopt least-privilege models for browser extensions and SaaS applications
Harden Supply Chains: Require artifact signing, verify dependencies, and monitor for trust anomalies
Eliminate Phishable Authentication: Enforce phishing-resistant MFA like FIDO2/WebAuthn as default

Fake Booking Emails Redirect Hotel Staff to Fake BSoD Pages Delivering DCRat

"We're chasing shiny new defense strategies while basic protections remain incomplete," Bustan concludes. "Attackers optimized the fundamentals. Our response must be perfecting those same fundamentals rather than reacting to hype."

For deeper analysis of emerging attack patterns, OX Security will host the webinar Threat Intelligence Update: What's Been Working for Hackers available here.