Check Point reports AI systems autonomously conducting live cyberattacks
Check Point Research's 2026 Annual AI Security Report documents a transition where AI systems have moved from assisting cyberattacks to autonomously operating them in live intrusions, affecting targets from state-sponsored campaigns to criminal breaches of Mexican government agencies. The report identifies that attackers are exploiting AI's agentic architecture through jailbreaks and planted configuration files, enabling the creation of malware and attack toolkits, while AI-powered services conduct phishing and voice-based social engineering at scale.
What happened
Check Point Research released its Annual AI Security Report 2026 on July 14, 2026, documenting the transition of AI from an assistive tool to an autonomous operator in live cyberattacks.
The report identifies AI conducting hands-on work inside live intrusions, from China-nexus espionage campaigns to criminal breaches of multiple Mexican government agencies.
AI systems are building deployment-ready malware and attack suites, including the VoidLink command-and-control framework created in under a week with 88,000 lines of code, and attackers are exploiting jailbroken mainstream commercial AI models through agentic architectures and planted configuration files.
Reported impact
Enterprises face shortened vulnerability windows giving defenders less time to identify and patch flaws, reshaping attack techniques and defensive priorities.
Virtual identity systems are no longer reliable trust anchors as AI enables forged voice, face, documents, and live video that are now widely used in multi-channel social engineering attacks.
Index analysis
Detections of malicious payloads increased sharply between March and May 2026, rising roughly fivefold and approaching 1% of observed prompts in May, indicating indirect prompt injection is becoming operationally relevant.
An AI-enabled criminal tooling market has matured with phishing-as-a-service kits embedding jailbroken language models and conversational AI voice-agent services running vishing and one-time-passcode theft at scale.
Governance controls
AI systems themselves present an expanding attack surface with vulnerabilities in their ability to separate data from instructions, ordinary software flaws in the surrounding stack, and supply-chain risks in a rapidly evolving ecosystem where security practices are not yet mature.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.