AI goes rogue: cybercrime and cyberwarfare enter the age of autonomous attacks

The initial phase of AI-driven cybercrime surfaced prominently in late 2025 with the discovery of PROMPTFLUX malware, which leveraged Google's Gemini API to dynamically obfuscate its behavior and evade security systems. This marked a pivotal shift as cybercriminals began integrating large language models and APIs to create self-modifying malware, effectively lowering the technical barriers that once confined sophisticated attacks to elite threat actors. As noted in November 2025, these AI-powered tools were already circulating in underground markets, signaling the dawn of more automated and scalable cybercrime accessible to mid-tier operators.

By late 2025, AI had become the indispensable '24/7 intern' for cybercriminals, automating mundane tasks such as chaining exploits, scripting attacks, and generating tailored phishing emails—capabilities previously exclusive to top-tier advanced persistent threats (APTs). This democratization accelerated the speed and scale of operations, effectively making human hackers up to four times faster. However, rather than fully autonomous hacking, AI primarily enhanced human operators by streamlining reconnaissance and attack preparation, thereby reshaping the cyber threat landscape and empowering mid-tier actors to execute complex campaigns with less expertise.

The emergence of malicious AI language models like WormGPT 4 and Kawaii GPT in December 2025 further lowered entry barriers by offering phishing, malware code generation, and automated reconnaissance tools without ethical constraints or safety filters. Priced affordably or even free, these models democratized sophisticated attack automation, enabling less skilled criminals to achieve meaningful success rates—estimated at 8-10%—across vast target pools. While experts like Ian Malloy viewed these tools as overhyped, their availability nonetheless streamlined operations for a broader range of attackers, marking a critical step in the widespread democratization of AI-powered cybercrime.

By early 2026, the professionalization and democratization of cybercrime had accelerated dramatically, with ransomware cartels like Dragon Force lowering entry barriers through online affiliate registration, test runs, and scripted ransom demands—transforming ransomware-as-a-service into a scalable, user-friendly ecosystem. Concurrently, darknet AI chatbots proliferated, increasing 219% year-over-year, and operated under reputation-based access controls that extended sophisticated AI capabilities beyond elite actors. Financially motivated groups such as Storm 1877 and large decentralized DPRK operations like Jasper Sleet rapidly integrated AI across workflows, enabling rapid iteration, autonomous malware development, and social engineering enhancements. This widespread adoption leveled the playing field, allowing traditionally less capable threat actors to deploy advanced, AI-generated malware and conduct complex, targeted attacks with unprecedented speed and scale.

By late 2025, the U.S. government had decisively ramped up investments in autonomous AI-driven offensive cyber capabilities, exemplified by a $12.6 million contract awarded to the stealth Arlington startup '20'. Staffed by veterans from cyber command and intelligence, '20' is pioneering tools capable of autonomously targeting and striking hundreds of systems simultaneously, signaling a strategic pivot from AI-assisted to fully autonomous cyberwarfare frameworks. This aggressive push mirrors a global trend, with adversaries like China also integrating AI agents into their cyber operations, underscoring a new era where state-sponsored cyber offense is rapidly evolving toward automation at scale.

By early 2026, industry leaders such as Kevin Mandia’s Armadin further crystallized this trajectory, aiming to build 'the ultimate attacker' through AI agents that automate cyber offense at the speed of compute. Mandia emphasized that attackers historically adopt new technologies before defenders, predicting a future within five years where autonomous AI-driven offense will train autonomous defense systems, potentially removing humans from the loop entirely. This vision reflects a profound military and state interest in developing nation-state grade offensive AI cyber capabilities to maintain strategic dominance in an increasingly automated battlefield.

Simultaneously, state-sponsored groups like North Korea’s Storm 1877 and Jasper Sleet exemplify how authoritarian regimes are rapidly integrating AI into cyber operations with startup-like agility. These groups leverage AI across reconnaissance, malware development, and social engineering, enabling rapid iteration and scaling of attacks. Jasper Sleet’s decentralized structure, with thousands of operators experimenting with diverse AI tools, contrasts with more bureaucratic intelligence units, highlighting how flexible organizational models accelerate early AI adoption in offensive cyber warfare.

Underlying these developments is a recognition that governments maintain a monopoly on offensive cyber capabilities, which now naturally extends into AI-driven operations to control and protect digital terrain. This strategic imperative drives states to collaborate with private industry—leveraging dual-use AI tools—to accelerate offensive capabilities before threats spiral beyond control. However, as offensive AI cyber tools become more autonomous and potent, legal authority and ethical frameworks become critical, since adversaries will inevitably exploit these technologies, making governance and understanding of these systems essential to maintaining strategic advantage.

By early 2026, AI had evolved from a mere assistive tool to a core driver of sophisticated, autonomous cyberattacks, exemplified by the VoidLink malware framework which was almost entirely generated by AI under a single developer using ByteDance's TRAE Solo assistant. This shift enabled lone actors to compress months of malware engineering into days, producing adaptive, cloud-aware malware written in less commonly monitored languages like Zig, thereby creating significant detection gaps and challenging traditional security paradigms.

Ransomware operations have professionalized and democratized with the integration of AI and cloud technologies, as groups like Dragon Force lowered entry barriers by offering affiliate programs and support services, enabling even inexperienced attackers to launch scalable, persistent attacks. The emergence of fully autonomous AI agents such as JadePuffer, which executed end-to-end ransomware campaigns exploiting known vulnerabilities and credential failures at machine speed without human intervention, marks a pivotal escalation in attack sophistication and operational autonomy.

Throughout 2026, AI-native offensive platforms like CyberStrikeAI and hackerbot-claw demonstrated the capability to conduct large-scale, multi-stage cyber campaigns with minimal human input, compromising hundreds of devices globally and orchestrating destructive supply-chain attacks. These developments reflect a broader trend where AI is embedded across the entire attack lifecycle—from reconnaissance and exploit development to phishing and infrastructure management—transforming cybercrime operations into agile, startup-like entities rapidly iterating and scaling their methods.

The rapid weaponization of AI has lowered the technical barriers for complex exploits, enabling attackers to autonomously discover and weaponize zero-day vulnerabilities at unprecedented speeds, as evidenced by Google's May 2026 report on the first AI-crafted zero-day targeting two-factor authentication in an open-source admin tool. Attackers now leverage premium and open-source AI models to automate account creation, exploit generation, and attack orchestration, forcing cybersecurity agencies like CISA to shorten remediation timelines and prioritize advanced defensive measures against this accelerating AI-driven threat landscape.

By late 2025 and into 2026, underground markets have rapidly commodified premium AI services, exemplified by malware like PROMPTFLUX leveraging Google's Gemini API for dynamic evasion, and open-source platforms such as CyberStrikeAI and hackerbot-claw that automate large-scale attacks with minimal expertise. These AI-native toolkits, integrating hundreds of security tools and capable of scanning tens of thousands of repositories, have lowered the barrier to entry for cybercrime, enabling even script kiddies and state-adjacent actors to conduct sophisticated campaigns across global targets, as seen in the CyberStrikeAI attacks on 600 FortiGate devices in 55 countries.

The illicit AI ecosystem extends beyond toolkits to a thriving black market for premium AI accounts, where paid subscriptions are stolen, bulk-created, or resold—often bundled and discounted—to circumvent regional sanctions and onboarding hurdles in countries like Russia, Iran, and North Korea. Flare analysts documented hundreds of Telegram listings offering these accounts, which empower a wide spectrum of threat actors from low-skilled hackers to extremist groups, democratizing access to advanced AI capabilities that were once the preserve of skilled developers.

AI-driven phishing-as-a-service kits, such as those marketed through Telegram for under $100 weekly, have revolutionized cybercrime by enabling mass production of highly customized scam sites using premium models like Gemini. Google's 2026 lawsuit against Outsider Enterprise revealed operations generating over one million fraudulent URLs and 2.5 million scam texts in mere weeks, resulting in $1.9 billion in losses, illustrating how AI commodification has exponentially scaled both the volume and sophistication of phishing attacks accessible to a broad attacker base.

Extremist groups like Boko Haram and Islamic State have institutionalized AI adoption by creating dedicated units that exploit AI chatbots for attack planning and weapons design, facilitated by the ease of bypassing AI safety guardrails. Simultaneously, cybercriminals such as the Russian-speaking 'bandcampro' have repurposed open-source AI tools like Google's Gemini CLI to autonomously manage botnets, automate command-and-control migrations in under six minutes, and perform password guessing and data dump analysis, demonstrating how AI commodification is enabling even unsophisticated actors to execute complex cyber operations with minimal resources.

By 2026, cybercrime has evolved into a fully industrialized, AI-driven enterprise, with autonomous agents executing complex campaigns at machine speed, as Trend Micro's Ryan Flores predicts. This transformation introduces unprecedented challenges for defenders, who must shift from reactive measures to proactive resilience strategies that integrate ethical AI use, adaptive defense, and continuous human oversight to keep pace with polymorphic malware, real-time adaptive intrusion campaigns, and AI-powered ransomware ecosystems capable of self-management and automated extortion.

The democratization of AI tools like WormGPT and Kawaii GPT has lowered the technical barrier for cybercriminals, enabling even low-skill actors to launch sophisticated phishing, malware generation, and reconnaissance attacks. While these tools currently augment less advanced threat actors, their subscription-based commercial models commoditize cybercrime infrastructure, expanding the threat landscape and complicating governance efforts, as defenders grapple with detection difficulties posed by indirect malicious prompts and creative evasion tactics such as prompt 'jailbreaking' through poetry.

The rapid automation and scale of AI-driven attacks, exemplified by cases like the autonomous JadePuffer ransomware and AI-crafted zero-day exploits thwarted by Google, compress detection and response windows from months to mere seconds. This acceleration demands dynamic security strategies including rigorous patch management, least-privilege controls, behavioral anomaly detection, and continuous monitoring of AI assistant permissions to counteract threats that exploit hybrid cloud environments, software supply chains, and post-authentication API vulnerabilities, while human expertise remains indispensable for interpreting AI outputs and coordinating rapid incident responses.

Governance and regulatory frameworks are struggling to keep pace with the dual-use nature of AI in cybersecurity, where the same AI capabilities can be harnessed for both defense and offense. The proliferation of open-source AI models that can be modified and weaponized independently of commercial providers, combined with the rise of AI-powered phishing-as-a-service operations generating billions in losses, underscores the urgent need for multi-stakeholder collaboration involving private sector, law enforcement, and policymakers. Despite the increasing automation of attacks, experts like Kevin Mandia and Google analysts emphasize that human expertise remains critical to managing AI tools effectively and preventing overreliance on adversarial technologies.

By late 2025, the geopolitical landscape of cyber warfare was rapidly evolving as nation-states like the US and China aggressively invested in AI-driven offensive cyber capabilities. The US-backed stealth startup 20 exemplifies this shift, aiming to automate strikes against hundreds of targets simultaneously, moving beyond AI-assisted human operators to fully autonomous offensive systems. This industrialization of AI-enabled cyber threats signals a new era where cyber offense is no longer a human-centric service but a scalable, automated operation, as highlighted by Trend Micro's prediction that 2026 will mark the full automation of cybercrime campaigns.

The integration of AI into cyber operations has escalated the sophistication and scale of attacks, with nation-state actors such as North Korea's Lazarus Group and Iran’s APT42 leveraging advanced AI tools like Google's Gemini for hyper-personalized phishing, autonomous malware development, and real-time adaptive payload generation. These developments undermine traditional detection methods and expand the attack surface to encompass the entire AI stack—from models and APIs to human trust—thereby intensifying the geopolitical arms race in AI-enabled cyber offense and necessitating robust global AI governance and security investments.

AI's role in cyber operations extends beyond offense to reshape national security strategies through accelerated vulnerability discovery and exploitation, as demonstrated by Anthropic's AI model uncovering hundreds of high-severity software flaws. This continuous competition among nations, including persistent US offensive and defensive cyber activities, underscores AI as a critical strategic asset in intelligence and defense, with milestones like DARPA's Cyber Grand Challenge and the AI Cyber Challenge marking significant advances in AI's cyber capabilities.

Beyond kinetic cyber attacks, AI is emerging as a potent tool for cognitive and disinformation warfare, enabling hyper-personalized manipulation that can fracture societies at the individual level. As Tony from X super observed in early 2026, AI-powered disinformation campaigns can turn neighbors against each other by subtly shaping personal realities, effectively weaponizing perception itself. This zero-cost cyber warfare, combined with geopolitical tensions exemplified by Iranian cyber operations amid ongoing conflicts, highlights the urgent need for proactive resilience strategies that integrate ethical AI governance and human oversight to safeguard democratic institutions and global stability.