Artificial-intelligence capabilities are rapidly reshaping the cyber-threat landscape, according to a series of reports and demonstrations released ahead of the Black Hat AI Summit in Las Vegas this week. CrowdStrike’s 2025 Threat Hunting Report says hostile actors are already “weaponizing and targeting AI at scale”, while Netskope Threat Labs warns that ungoverned “Shadow AI” tools inside enterprises are expanding attack surfaces faster than security teams can lock them down. The warnings were underscored by university researchers who showed that an AI agent could, without human intervention, replicate a well-known multi-stage breach—scanning a network, exploiting a vulnerability, installing malware and exfiltrating data. The experiment, published 4 August, demonstrates that large-language-model ‘brains’ can now orchestrate specialised AI sub-agents to run end-to-end operations that once required skilled human attackers. Vendors are rushing to respond. GTB Technologies, Skyflow and AirMDR each unveiled products that embed generative-AI detection and automated response into data-loss prevention, cloud security and managed SOC platforms, respectively. Separately, a survey by Enterprise Management Associates ranked Keeper Security’s cloud-native privileged-access platform highest for ease of deployment and customer satisfaction, highlighting demand for zero-trust controls that can limit what compromised AI agents can reach inside corporate networks. With generative models diffusing across both offense and defense, security specialists say organisations should tighten governance of internal AI projects, adopt zero-trust access policies and run adversarial simulations to understand how autonomous systems might be turned against them.