Cloud attacks surged by 136% in the last six months, with North Korean agents infiltrating 320 firms via AI-generated personas. The group known as Scattered Spider now launches ransomware attacks in under 24 hours. At Black Hat 2025, industry experts showcased a viable solution: agentic AI, which delivers tangible results rather than mere promises. CrowdStrike recently uncovered 28 North Korean agents working as remote IT staff, highlighting how agentic AI is transitioning from theory to effective threat detection. Most vendors at Black Hat 2025 demonstrated readiness for action, emphasizing operational capacity over theoretical claims. CISOs consulted by VentureBeat noted significant improvements in processing alerts and reducing investigation times, though specific gains vary based on implementation maturity.
AI Scaling Hits Its Limits
Power restrictions, rising token costs, and processing delays are impacting enterprise AI. Discover strategies from leading teams, such as turning energy use into a strategic edge, designing efficient systems for real throughput improvements, and achieving competitive ROI with sustainable AI systems. Secure your spot to stay ahead: https://bit.ly/4mwGngO
VentureBeat observes that security teams are experiencing practical efficiency improvements, such as reduced mean time to investigate (MTTI), enhanced threat detection rates, and better resource use. Black Hat 2025 marked an inflection point where the focus shifted from AI’s potential to its concrete impact on security operations.
The agentic AI arms race shifts from promises to production
Agentic AI was a central theme at Black Hat 2025, with many sessions discussing the ease with which attackers compromise agents. VentureBeat noted over 100 announcements of new agentic AI applications, platforms, or services. Vendors are emphasizing real use cases and outcomes over empty promises. This shift reflects an urgency to close hype gaps and deliver results. CrowdStrike’s Adam Meyers explained the push for this shift during an interview with VentureBeat, emphasizing the platform’s function for building automations. VentureBeat highlights the scope of the threat and the rapid pace of ransomware deployment, which demands immediate responses and human-in-the-loop threat hunting. Meyers revealed that last year’s operations included 60 billion hunting leads, resulting in significant escalations and emails sent to customers. Microsoft Security introduced autonomous investigative capabilities in its Security Copilot, while Palo Alto Networks showcased Cortex XSOAR’s new agentic capabilities for autonomous alert triage.
Cisco’s major Black Hat announcement involved its release of Foundation-sec-8B-Instruct, a conversational AI model for cybersecurity, outperforming larger models on security tasks using a single GPU. This model is notable for its fully open-source nature, accessible for on-premises deployment without vendor lock-in. SentinelOne highlighted Purple AI’s proactive capabilities to predict adversary actions and adjust defenses.
FAMOUS CHOLLIMA infiltrated 320 companies last year, signaling a sharp increase in security threats. Adversaries employ AI throughout their operations, from creating false identities to conducting interviews and performing tasks. Supporting infrastructure, such as Arizona-based facilitators maintaining 90 laptops, illustrates the sophistication of these operations. CrowdStrike identified numerous malicious insiders among AI-enhanced operators. Despite technological advances, human insight remains vital in threat detection. Vendors agree that agentic AI complements rather than replaces human analysts. Splunk’s Mission Control showcases this collaboration by automating routine tasks, while humans handle complex decisions.
Emerging challenges include AI potentially becoming the next insider threat, prompting discussions on standardization and governance. The Cloud Security Alliance announced initiatives focused on agentic AI security standards, while vendors committed to collaborative efforts for AI interoperability.
Bottom Line
Black Hat 2025 confirmed the rising impact of AI-driven attacks across unexpected surfaces, such as human resources and hiring processes. FAMOUS CHOLLIMA’s infiltration threatens to divert resources to North Korea’s weapons programs while stealing valuable IP. The balance of securing core IP, national security, and customer trust hangs in the accurate management of these threats.