AI-assisted cyberattacks exploiting zero-day vulnerabilities before patches exist reached a record scale in the first quarter of 2026, according to Mandiant's M-Trends 2026 report, which found that 28.3 percent of all tracked CVEs were exploited within 24 hours of public disclosure — a rate that security teams described as operationally unmanageable using conventional defense methods.
The Time-to-Exploit Has Gone Negative
The most alarming finding in the Mandiant report is not the volume of attacks — which researchers expected to rise — but the timing. Across a dataset of hundreds of high-severity vulnerabilities tracked through March 2026, the firm found that the average time between public disclosure and first observed exploitation had effectively collapsed. In a growing subset of cases, exploitation was detected before vendors had issued patches at all, making the standard advice to "patch quickly" meaningless as a primary defense.
"The window is gone," said a principal threat intelligence analyst at a major financial services firm in Charlotte, North Carolina, who reviewed the Mandiant findings before their public release. "We used to think in hours. Now we're thinking in minutes — and sometimes the attack is happening before we even know there's a hole."
IBM's parallel threat intelligence division reached similar conclusions. Its analysis, released last month, identified ransomware, spear-phishing, and data theft operations as the three most prevalent attack categories in 2026, with AI-generated tooling playing a significant role in scaling each. In phishing campaigns specifically, IBM found that AI-generated lures were producing click-through rates nearly three times higher than manually crafted messages from just three years ago.
Artificial Intelligence Shifts the Offense-Defense Balance
The AI factor represents a genuine structural shift in the threat landscape, one that does not fit neatly into traditional security frameworks built around signature detection, patching cadences, and human analyst review cycles. AI enables attackers to iterate faster, to personalize attacks at scale, and to identify exploitable conditions in large, complex codebases in a fraction of the time it takes human researchers to complete the same analysis.
The Hacker News, tracking threat actor activity through the first five months of 2026, described the year as the emergence of "AI-assisted attack pipelines" — not a single AI tool but integrated chains of purpose-built models handling reconnaissance, payload generation, and evasion optimization as a coordinated workflow. Nation-state actors linked to Russia, China, and North Korea have been the fastest adopters, but criminal ransomware organizations are closing the gap quickly.
The development contributed directly to the Trump administration's decision to issue an executive order last week directing federal agencies to develop AI-specific cybersecurity benchmarks and create a centralized AI Cybersecurity Clearinghouse. The order also asks AI developers to voluntarily submit powerful new models for government review before public release — a provision the White House said was partly motivated by evidence that frontier AI systems had dramatically lowered the barrier to sophisticated cyberattacks.
The Defense Challenge
The practical implication for security teams is a resource problem as much as a technical one. Keeping pace with a threat environment where exploitation can precede disclosure requires continuous monitoring, behavioral detection, and network segmentation that most mid-sized organizations have not implemented at the necessary depth. The talent gap is severe: the industry estimates a shortfall of more than 3 million qualified security professionals globally, a figure that has not improved meaningfully in three years.
Defenders are turning to their own AI tools — automated vulnerability scanning, AI-assisted triage, and large language models trained to identify anomalous network behavior — but the adoption curve lags the threat by a wide margin. In a survey of 800 enterprise security leaders conducted by a San Francisco-based research firm in April, 62 percent said their AI-powered defense tools were still in pilot or limited deployment, while 41 percent said they had experienced a material security incident attributable to an AI-assisted attack in the preceding 90 days.
The divergence in adoption speed is the core of the problem. Attackers face no organizational procurement cycles, no board approvals, no vendor evaluation processes. A criminal group in Eastern Europe can deploy a new AI-powered attack tool in days. A Fortune 500 company's security team in Houston or Seattle might take 18 months to move the same capability from pilot to production. That asymmetry is not new, but AI has widened it considerably.
What Comes Next
Analysts tracking the threat landscape say the second half of 2026 will be defined by two converging pressures: the continued proliferation of AI offensive tooling into the broader criminal ecosystem, and the first serious efforts by governments to establish international norms around its use. The Biden administration's AI safety frameworks were largely voluntary; the Trump order pushes further into the technical verification space, but enforcement mechanisms remain unclear and largely untested.
What is clear is that the environment Mandiant describes in its 2026 report — one in which the window for defense has closed before most organizations know it has opened — is not a temporary condition. It is the new baseline. Security teams in every sector, in every city, will need to be operating continuously and adaptively or they will be operating reactively. And reactive, as Mandiant's data shows, is already too late.