AI broke the cybersecurity world. What’s next?

A wave of automated intrusions across cloud infrastructure, developer platforms, and financial systems has forced technology companies into sustained emergency response. As detailed in The Verge's report, cybersecurity leaders warn that open-weight artificial intelligence models paired with publicly available exploitation frameworks are fundamentally overturning corporate defenses. Security teams faced major bugs in PHP and Node alongside chained enterprise intrusions throughout October and into November. Alex Stamos, chief security officer at Cognition and lecturer at the Stanford University Cyber Policy Center, warned that the multibillion-dollar cybersecurity industry is overdue for disruption as both private criminal crews and state-backed actors from Russia and North Korea adopt automated attack pipelines.

This reporting examines the technical mechanics of the current intrusion wave, the specific toolchains enabling automated breach chains, and the structural implications for enterprise security teams. Engineering organizations, cloud infrastructure providers, and software developers face an environment where attackers no longer rely on commercial frontier models with safety logs or manual exploit development. By examining how recent intrusions compromised providers such as Fireworks and Cursor before cascading to thousands of downstream customers, this account provides security professionals with factual insight into automated exploitation workflows, uncensored open-weight models, and defensive remediation practices.

AI broke the cybersecurity world. What’s next: what actually changed

Attack capabilities reached an inflection point where adversaries now automate nearly the entire military-derived cyber kill chain. Traditionally, threat actors spent weeks or months conducting manual vulnerability research, writing exploit code, and staging human operators to deploy payloads against selected enterprises or government contractors like Lockheed Martin. That operational sequence has inverted because automated software engines inspect target infrastructure, detect vulnerabilities in real time, and generate operational exploit code without waiting for human intervention. The speed of discovery eliminates the defensive response window that security teams previously used to patch known weaknesses.

The widespread availability of uncensored open-weight models removed the barriers that previously limited automated attacks to well-funded intelligence services. Rather than paying for commercial frontier models that enforce safety guardrails and preserve query logs for law enforcement, adversaries run localized machine learning models directly on consumer workstations. Stamos demonstrated that running a quantized, obliterated GLM-5.3 model locally on an M5 Mac Ultra Studio allows individuals to run offensive tooling independently. This transition mirrors the early internet frontier of the 1990s, when small underground groups found widespread vulnerabilities, but modern systems expose connected global commerce rather than isolated academic networks.

AI broke the cybersecurity world. What’s next: how it works

The operational pipeline driving these attacks relies on publicly accessible exploit frameworks driven by local model inference. Attackers deploy an open-source Chinese orchestration framework named ARTEX, which is publicly accessible with an English translation hosted on GitHub. When engineers or threat groups connect ARTEX to an uncensored back-end model such as obliterated GLM-5.3, the system scans target networks, maps unpatched weaknesses, and writes custom payloads designed specifically for the target environment. The engine executes these steps continuously without requiring external network calls to vendor-hosted application programming interfaces.

Recent campaigns demonstrate how automated tools chain individual intrusions together into catastrophic supply-chain compromises. Attackers first breached Fireworks, an artificial intelligence infrastructure service used by numerous technology companies, and extracted stored API credentials. Using those compromised keys, the attackers infiltrated the developer platform Cursor and published a malicious plugin that was distributed to 5,000 corporate environments, opening thousands of secondary entry points. This systematic credential harvesting and automatic downstream distribution creates a rolling, self-propagating breach pattern across interconnected cloud services.

AI broke the cybersecurity world. What’s next: why it matters

The economics of digital defense have broken down because attacking enterprise software is now substantially cheaper than defending it. For decades, the software sector maintained relative parity by offering bug bounties, hiring security researchers from conferences like DEFCON, and establishing responsible disclosure norms. Today, DEFCON hosts between 40,000 and 60,000 attendees, while criminal syndicates operating in jurisdictions outside Western legal reach utilize automated tools with strong financial incentives to breach targets. A single open-source download now grants untrusted actors offensive capabilities that previously required dedicated nation-state research divisions.

The financial scale of modern incidents underscores the systemic exposure created by automated tooling. Attackers recently executed the Binance hack, which stands as the largest theft of financial assets in recorded human history. Security executives report that their internal defense teams have operated at the equivalent of DEFCON 2 throughout October and early November to manage simultaneous intrusions across multiple service providers. The rapid succession of breaches across specialized AI hosting providers and infrastructure neoclouds demonstrates that existing commercial defense suites cannot keep pace with automated real-time exploit generation.

AI broke the cybersecurity world. What’s next: who is affected

The primary victims of this intrusion surge span core developer tooling, major financial institutions, and emerging cloud providers. The automated deployment of ARTEX powered by open-weight models first surfaced publicly against three South Korean commercial banks, which suffered automated system breaches before defensive teams could intervene. Following those financial breaches, roughly 200 additional organizations were hit within a single seven-day period, illustrating how quickly automated pipelines identify and exploit common technical weaknesses across disparate industries.

Downstream software companies and cloud service customers are particularly vulnerable through integrated vendor pipelines. The 5,000 corporate customers that received the rogue Cursor plugin illustrate how quickly supply-chain compromises propagate through developer workstations into broader corporate networks. Even companies with dedicated security personnel face severe risk because third-party infrastructure components, open-source libraries, and machine learning platforms carry shared credentials. Cognition has dispatched internal engineering teams to assist compromised businesses in resolving underlying software flaws, confirming that even well-resourced engineering firms are finding their perimeter protections penetrated by automated tooling.

AI broke the cybersecurity world. What’s next: what to watch

Industry observers are monitoring the emergence of automated defensive software capable of matching the speed of autonomous offensive agents. Because Cognition and peer technology companies develop automated programming tools, software providers are experimenting with using code-generation agents to identify and patch system vulnerabilities before external scanners locate them. Defensive teams must transition from manual incident response workflows to automated code remediation, as human analysts cannot triage thousands of simultaneous real-time intrusion attempts across enterprise environments.

Government regulators and international intelligence bodies will also increase their focus on open-source code repositories and the distribution of uncensored model weights. As tools like ARTEX circulate openly on platforms like GitHub, policymakers face difficult decisions regarding open-weight model releases that can be stripped of safety tuning. Security leaders anticipate an uncomfortable transition period characterized by sustained operational friction until organizations overhaul legacy network architectures, isolate sensitive developer environments, and deploy autonomous agents capable of repairing breached codebases at machine speed.