The Industrialization of Deception: AI-Driven Threats

By Emma DeCarli · September 22, 2026

The line between kinetic warfare and digital sabotage officially blurred in the aftermath of the early 2026 campaigns in Iran. While the United States military campaign, "Operation Epic Fury," targeted traditional infrastructure, a more dangerous, global threat was simultaneously emerging for businesses: The era of autonomous breach. While the world watched the Strait of Hormuz, adversaries are quietly shifting their focus. They are no longer interested in "breaking in" through technical exploits; they are simply "logging in" by simultaneously weaponizing trusted relationships, service providers, and the machine-speed capabilities of artificial intelligence (AI). In this new era of geopolitical volatility, your greatest vulnerability isn't a line of code—it’s the exponential speed at which AI can dismantle your most fundamental human-trust protocols.

This blog is a part of a three-part series. If you haven’t already, check out part 1 and part 2 in this series for more information.

Why this matters to you

The primary implication of this strategic shift is the immediate and catastrophic failure of reputation-based security models. Since threat actors are now leveraging AI to orchestrate machine-speed, self-directed campaigns that seamlessly mimic legitimate behavior, traditional Indicators of Compromise (IoCs) are rendered obsolete. For executive stakeholders, the "so what" is a clear mandate: Security must transition immediately from a static, compliance-based posture to a model of continuous readiness. This new framework prioritizes the continuous verification of identity integrity and behavioral intelligence over outdated perimeter defense strategies.

Anthropic Claude Code: The arrival of autonomous espionage

Background: A significant cyber espionage event, the GTG-1002 campaign, ran from late 2025 into 2026 and featured operations notable for their highly autonomous, AI-driven nature. This state-sponsored attack, using the Claude Code tool and agentic orchestration, had AI agents handle 80% to 90% of all operational tasks. Human operators were involved in only 4-6 critical decisions, like target selection and final data exfiltration. AI managed the tactical execution of the middle four phases of the six-phase operation.

This campaign is attributed to the state-sponsored Chinese threat actor, designated GTG-1002 (PRC-nexus). This high level of automation signifies a significant advancement in cyber warfare, effectively making elite-level offensive capabilities more broadly accessible.¹

Strategic 'so what': The agentic shift is defined by the transition from human-paced incidents to self-directed, machine-speed campaigns. Because the AI operates 24/7 without rest and at a speed impossible to match for human hackers, traditional detection windows are now obsolete. The strategic risk is that commercially available AI can now shoulder the labor of a top-tier espionage group.

Actionable insight: AI excels at finding the path of least resistance. Protection against high-speed, AI-driven threats necessitates a shift from reactive blocking to a strategy of verification and continuous exposure management. This involves replacing bypassable security measures, such as SMS-based multi-factor authentication (MFA), with phishing-resistant hardware passkeys and implementing mandatory out-of-band (OOB) human verification. Traditional quarterly vulnerability scans are obsolete; continuous exposure monitoring needs to match the 24/7 cadence of AI-powered scans. This can help the organization break the attacker’s automated chain and adopt an active resilience posture.

PromptSpy: Adaptive AI-driven mobile persistence

Background: Identified earlier this year (February 2026), PromptSpy is the first known Android malware to use generative AI (Google Gemini) to maintain its presence on a device. Leveraging accessibility services, it analyzes the user's screen in real time and uses AI to generate instructions to interact with the device's interface, ensuring the malicious app remains "pinned."

While linked to Chinese threat actors, the operation is characterized by its use of Argentina-based infrastructure for executing financial fraud. The actor uses trusted tools like Google Gemini to mimic normal automation assistant behavior. The malware captures lockscreen data, device info, and records screen activity as video. It blocks uninstallation via invisible overlays, forcing users to perform a hardware-level "safe mode" reboot to remove the threat.²

Strategic 'so what': PromptSpy represents an evolution toward adaptive tradecraft, moving beyond static code. This is because the malware interprets on-screen elements instead of relying on hard-coded coordinates, allowing it to adjust to any security update or device layout. This thins the line between helpful automation and malicious intent, as the malware uses the same AI tools the corporation might be deploying for productivity.

Actionable insight: Removal of PromptSpy necessitates a hardware-level intervention. Due to the malware's use of invisible overlays to prevent uninstallation, users must reboot their device into Safe Mode to successfully remove it. Furthermore, mobile security policies should enforce strict limitations on the use of accessibility services.

The deepfake heist: $25M transactional impersonation

Background: In early 2026, a high-profile industrial-scale deepfake scam resulted in a $25 million loss for a built-environment engineering firm. Attackers used AI-generated video and audio to impersonate multiple company leaders during a video conference, convincing a finance employee to authorize a massive fraudulent transfer.

This heist was conducted by a professionalized industrial-scale fraud gang specializing in multi-channel social engineering. These groups are increasingly using deepfake-as-a-service platforms on the dark web to impersonate C-suite executives with near-perfect voice and video cloning.³

Strategic 'so what': This is industrial-scale impersonation for profit. The "Deepfake CEO" has become a standard tool for heists, exploiting organizational hierarchies where employees are conditioned to say "yes" to leadership. The strategic concern is that voice and video are no longer reliable proofs of identity; a fabricated recording of an executive could circulate and destroy brand value before it can be disproven.

Actionable insight: Formulate zero-trust human protocols. High-value transactions must require multi-channel, out-of-band verification (e.g., an analog callback or a pre-shared code) that does not rely solely on digital signals.

Double-tap skimming: The weaponization of technical "glitches"

Background: Earlier this year, in February 2026, a significant campaign of double-tap skimming targeted global supermarket chains. The method involves attackers initially injecting a fraudulent payment form. After a user submits their information to this fake form, the form intentionally fails or refreshes, prompting the user to re-enter their payment details, this time into the legitimate payment form.

This campaign is linked to the s1ngularity cluster, which is a sophisticated actor notorious for credential mutualization and deploying supply chain worms. Their strategy involves stealing credentials (such as GitHub tokens) from one victim to establish the infrastructure needed to infiltrate and exfiltrate data from the next. Although their national origin remains hidden, this actor is known for weaponizing trusted AI tools to automate both reconnaissance and data theft.⁴

Strategic 'so what': This attack targets the core of consumer trust by mimicking common technical glitches. It also represents a brand-poisoning event; if consumers feel a retailer's digital checkout is glitchy or unsafe, they will migrate to competitors. The fraud is nearly indistinguishable from legitimate errors, allowing for silent, large-scale harvesting of payment data.

Actionable insight: Implement synthetic transaction monitoring: a proactive performance-testing method that uses automated scripts to simulate user interactions, ensuring your application’s critical paths are functional and fast before real users encounter issues. Regular audits of e-commerce applications should be considered a baseline requirement.

LunaLock: Intelligent AI-driven extortion

Background: Emerging at the start of January 2026, LunaLock represents an evolution in ransomware. Instead of simple mass encryption, the tool uses AI to scan stolen datasets and automatically identify the victim's most sensitive documents (M&A strategies or legal notes) to exert maximum pressure.

The LunaLock ransomware group is a next-generation syndicate that has shifted from volume to intelligence. The threat actor compromised over 95,000 user accounts associated with the Artists&Clients digital marketplace. LunaLock has escalated the threat by claiming they will poison the professional futures of victims by using technology to launder the stolen intellectual property.⁵

Strategic 'so what': Extortion is now a clinical and automated process. Attackers utilize AI to analyze stolen data more quickly than internal legal teams, maximizing leverage and making paying the ransom seem unavoidable.

Actionable insight: Business continuity must move beyond system restoration to data exposure management. Prioritize document-level encryption for your most sensitive intellectual property and strategic plans.

Conclusion: The age of verification

With early 2026's geopolitical conflicts serving as a catalyst for a new era of cyber warfare, the luxury of "good enough" security has expired. The 2026 threat landscape is no longer defined by the lone hacker, but by machine-speed, adaptive adversaries who have successfully industrialized deception. By merging the collapse of digital trust with the industrialization of deception, they are now leveraging AI to orchestrate autonomous campaigns that weaponize identities, bypass technical filters, and utilize trusted SaaS connectivity as a silent delivery system.

The strategic "so what" is undeniable: Compliance is not security. A green checkmark on an audit will not stop a $25 million deepfake heist, nor can it halt the self-directed, machine-speed autonomous espionage seen in the GTG-1002 campaign. The fundamental risk is now the complete breakdown of verification protocols.

The era of static, perimeter-based defense is over. To reclaim control, we must operationalize the four strategic pillars of continuous readiness—shifting our focus entirely from blocking threats at the edge to validating every action at the core. Our survival hinges on embracing an identity-first future, where we stop trusting the login and start verifying the intent.

The primary point of vulnerability is no longer a flaw in programming, but rather the reliance placed upon one's digital ecosystem.

¹ https://www.anthropic.com/news/disrupting-AI-espionage
² https://thehackernews.com/2026/02/weekly-recap-double-tap-skimmers.html
³ https://astt.net.au/2026/02/15/the-deepfake-ceo-scam-why-voice-cloning-is-the-new-business-email-compromise-bec/
⁴ https://thehackernews.com/2026/02/weekly-recap-double-tap-skimmers.html
⁵ https://simplysecuregroup.com/lunalock-ransomware-attacking-artists-to-steal-and-encrypt-data

Discover the latest cybersecurity research from the Trellix Advanced Research Center.