AI-Driven Cyberattacks Accelerate, Forcing Tech Giants and Governments to Deploy AI Defenses

The growing use of artificial intelligence in cyberattacks is creating faster, more adaptive threats, prompting a major push by technology companies and governments to develop AI-powered defenses and restrict access to their most powerful models. As these new attacks target critical infrastructure, companies like Google are giving select defenders early access to frontier AI systems, betting that the best way to fight malicious AI is with even more capable AI.

According to industry and government reports, AI is rapidly becoming a standard tool for cybercriminals. As of early 2026, one in four malicious data breaches was AI-enabled, a 56% increase from the previous year, while a staggering 86% of phishing attacks were driven by AI. The use of deepfakes in fraud attempts has also surged, accounting for roughly 6.5% of cases, up from about one in a thousand just three years prior. In 2025 alone, the FBI reported $893 million in losses from 22,364 complaints related to AI-driven fraud.

A New Breed of Cyberattack

The core change brought by AI is the rise of autonomous attacks that can operate at a speed and scale previously impossible. These systems, sometimes called "agentic AI attacks," can plan and execute malicious campaigns with minimal human oversight. An AI agent can reason about the results of its actions and decide on the next step, allowing it to continuously adapt its strategy. This capability is being used to target everything from financial systems and hospitals to power grids and water supplies.

AI's most significant impact has been in shortening the time between the discovery of a software vulnerability and its exploitation. Attackers are using AI tools to accelerate the weaponization of publicly disclosed flaws, known as n-day vulnerabilities. According to Google's Threat Intelligence Group (GTIG), the median time from a vulnerability's public disclosure to its confirmed exploitation fell from 120 days in 2025 to just 80 days in the first half of 2026, a drop potentially driven by AI agents. GTIG reported that 141 flaws were exploited between January and August 2026, surpassing the 127 exploited in all of 2025.

In some cases, AI agents are discovering previously unknown critical vulnerabilities. Google reported that one such AI discovery exposed sensitive personal information across healthcare software used by hospitals worldwide, though the company did not name the specific software affected. The most advanced attacks can be launched within 24 hours of a vulnerability's public disclosure, giving defenders a perilously small window to react.

Building an AI-Powered Defense

In response, leading tech companies are developing what they call AI-native cybersecurity defenses. The strategy involves using their most advanced models to find and fix vulnerabilities before attackers can, but it comes with a significant risk: the same tools could be turned into powerful weapons if they fall into the wrong hands. This has led to a highly cautious and restricted rollout of new defensive AI systems.

Google's Guarded Approach

On September 30, 2026, Google introduced Gemini 4 Argon, its latest frontier AI model, and began making it available to a select group of trusted partners through its Fairwind Program. The program, which launched earlier in the month, grants high-priority defenders—including governments, healthcare providers, and telecommunications companies—access to advanced AI to help them build better security before new threats emerge.

Google states that Gemini 4 Argon delivers "frontier performance" and shows "impressive leaps" in discovering software vulnerabilities compared to its predecessor. The model can autonomously find, validate, and even patch critical flaws. It excels at identifying potential attack surfaces and generating proof-of-concepts to confirm a vulnerability's existence.

Because of its powerful capabilities, Google is restricting access to Argon. A version of the model without its usual cyber guardrails is being released only to Google's internal teams and trusted external defenders. This allows them to use its full potential for defensive tasks but prevents its immediate misuse by malicious actors. Before any wider release, the company is working to strengthen safeguards against the model being used for unintended purposes. These "misalignment mitigations" include systems that monitor Argon's chain-of-thought and actions, automatically halting its execution if it deviates toward a malicious or harmful goal.

Organizations participating in the Fairwind Program must agree to strict governance terms. These include requiring user-level authentication, phishing-resistant multi-factor authentication (MFA), and tight access controls. Access to Argon is limited to internal cybersecurity, incident response, or penetration testing teams, and all employee use must be tracked.

A Global Mobilization

The shift toward AI-driven cyber warfare has prompted a broader response from industry and government. More than 100 companies and organizations, including OpenAI, Anthropic, Microsoft, and Amazon Web Services, have signed an open letter calling for a global effort to improve cybersecurity. They warn of a narrow "defenders' window" to strengthen security before AI-powered attacks become overwhelmingly sophisticated.

Governments are also taking action. The "Five Eyes" intelligence alliance—comprising Australia, the United States, the United Kingdom, Canada, and New Zealand—issued a joint warning about the threat. The alliance urged governments to coordinate defense across borders, fund the protection of essential services, and work to impose costs on attackers. While acknowledging AI's potential as a weapon, the group also stressed its power as a defensive tool.

The U.S. government is actively supporting the development of AI to improve national cyber resiliency and is enabling access to advanced AI models for defenders protecting critical infrastructure. The Department of the Treasury has been designated as the lead for a new government clearinghouse for AI cybersecurity tools. This concerted effort is reflected in market forecasts; the global AI in cybersecurity market reached $29.64 billion in 2025 and is projected to grow to $167.77 billion by 2035.