CrowdStrike Says AI-Triggered Leads Are Up 2.5x. That Counts Its Own Product, Not Attackers.

The 2026 Threat Hunting Report lands with a hard, checkable number — 88% of exploitation with public proof-of-concept code happens inside 48 hours. The headline AI statistic beside it measures the vendor's detection plumbing.

Aug 3, 2026Source: CrowdStrike

An illustration of two humanoid robots, one red and one black, squaring up in boxing stances.

CrowdStrike published its 2026 Threat Hunting Report on August 3, drawn from its own intelligence and OverWatch telemetry for the first half of the year. Adam Meyers, who runs counter-adversary operations, framed it as "AI is both the weapon and the target."

The number that survives scrutiny

Where a public proof-of-concept exists, 88% of the exploitation CrowdStrike saw happened within 48 hours of publication. Two China-linked groups, VAULT PANDA and GENESIS PANDA, moved on one critical web-application flaw inside 24 hours of disclosure. On the React2Shell campaign, OverWatch handled 800-plus hunting leads across more than 80 victims in four days. The supply-chain figures are equally concrete: 87% of software-registry threats in the first half involved npm packages, and North Korea-linked actors poisoned 131 trusted AI framework packages and more than 300 dependencies in a single day.

The number that does not mean what it says

The line being reported as evidence that AI-driven attacks grew 2.5x says something else entirely: agent-triggered detection now surfaces 2.5 times more threat leads than manually driven activity. Both sides of that ratio are CrowdStrike's. It describes how noisy its own agentic hunting has become. The company shipped Falcon protection for Copilot Studio agents and Claude Code four days before publishing.

Check the denominator

The 88% is 88% of exploitation CrowdStrike observed, of vulnerabilities that already had public exploit code — not of all vulnerabilities, and not a population anyone outside its install base can audit. No methodology statement accompanies the post. A "89% surge in machine-assisted attack activity" quoted in secondary coverage does not appear on CrowdStrike's own report page.

What LLMJacking actually is

One campaign fired nearly 200,000 model requests in two minutes. That is stolen enterprise credentials being drained for compute — resource theft, routinely rewritten as an AI-powered attack.

Why It Matters

This is the flagship annual report from the largest endpoint-security vendor, and it will set the AI-in-attacks narrative for the rest of 2026. The 48-hour exploitation window is the operationally useful figure; the AI statistic sitting beside it is a product metric, and the gap between the two is unusually easy to check.

Filed underAI Safety AI Agents AI Tools #AI security #CrowdStrike #npm #supply chain #threat intelligence