The head start defenders get on a newly published exploit has all but disappeared. In its 2026 Threat Hunting Report, CrowdStrike's Counter Adversary Operations team reports that from January through June 2026, 88% of CrowdStrike-observed exploitation of vulnerabilities with a public proof of concept was conducted within 48 hours of the PoC's release.
A proof of concept, or PoC, is demonstration code a researcher publishes to prove a bug is real. It has long been treated as a starting gun for attackers. What has changed is how fast they now run.
Hours, not weeks
China-nexus adversaries moved fastest of all. CrowdStrike says VAULT PANDA and GENESIS PANDA launched deliberate attacks within 24 hours of the public disclosure of a critical web application vulnerability. After the React2Shell disclosure, the company's OverWatch hunting team responded to more than 800 hunting leads across over 80 victims in four days.
The report offers CVE-2026-31431, a Linux local privilege escalation flaw, as a timeline in miniature. It was publicly disclosed on 29 April 2026, and an industry researcher released a PoC exploit and technical details the same day. The next day OverWatch saw widespread deployment of the exploit, with roughly 94% of events in the first 24 hours amounting to testing behavior based on the public PoC code. Belarus-nexus adversary UMBRAL BISON was found using it just over 20 hours after disclosure. CrowdStrike expects these timelines to keep shrinking, and judges that frontier AI models are likely to compress them further by speeding up vulnerability discovery and exploit development.
AI as both weapon and target
The second theme is AI systems themselves becoming the ground being fought over. CrowdStrike describes one LLMJacking campaign, meaning the theft and resale of access to someone else's AI model subscriptions, that generated nearly 200,000 API requests in two minutes, with what the company calls large-scale financial and operational impact.
DPRK-linked FAMOUS CHOLLIMA weaponized trusted AI-centric environments to compromise cryptocurrency and blockchain companies, in what CrowdStrike calls among the most sophisticated examples it has seen of the MITRE ATLAS technique AI Supply Chain Compromise. Defenders are also getting noisier data to sift: OverWatch found that AI agent-triggered detection leads now surface 2.5 times more threat leads than manually driven activity, making malicious behavior harder to separate from expected AI activity. We reported on a related problem in July when CrowdStrike catalogued the ways attackers hijack AI agents with hidden prompts.
The developer supply chain
Package registries remain the softest route into many organizations. CrowdStrike says 87% of identified software registry threats in the first half of 2026 involved npm packages. Financially motivated ALTERED SPIDER compromised more than 300 software dependencies in a single day, harvested credentials and pivoted into cloud environments. DPRK-nexus STARDUST CHOLLIMA used stolen maintainer credentials in March 2026 to compromise the Axios npm package and deliver its ZshBucket malware, then in June injected a malicious npm package as a dependency into at least 131 Mastra AI framework packages. Our coverage of a GitHub Actions flaw used to plant npm backdoors shows the same pattern from the CI/CD side.
Phones still work
None of this has displaced the telephone. Vishing intrusions in the first half of 2026 increased twofold compared with the second half of 2025, and monthly device code phishing attempts jumped 15 times over the past six months. CrowdStrike says eCrime actors CORDIAL SPIDER and SNARKY SPIDER used vishing to exfiltrate data from SaaS applications and compromise single sign-on accounts, with SNARKY SPIDER moving from account takeover to data theft in under five minutes.
What you should do
Treat the publication of a PoC as a countdown rather than a notification, and pull emergency patch windows for internet-facing systems down to hours. Watch AI service credentials and API spend for the anomalous spikes that betray LLMJacking. Pin and verify dependencies, and enforce phishing-resistant multi-factor authentication on package maintainer accounts. Finally, rehearse the help desk identity checks that vishing crews are built to defeat, and monitor device code authentication flows, which are easy to forget and increasingly abused. CrowdStrike now tracks more than 290 named adversaries.
This briefing is provided by IntelFusions for informational and defensive purposes only. It is based on sources assessed to be reliable at the time of writing, and analytic judgments carry the confidence levels indicated. Indicators of compromise are defanged; re-arm them only in controlled environments. IntelFusions is not affiliated with the organizations named and makes no warranty as to completeness or accuracy.