We are thrilled to announce the AI Cyber Model Arena, featuring a benchmark suite that includes 257 real-world challenges across five critical offensive domains: zero-day discovery, CVE detection, API security, web security, and cloud security. This initiative enhances everyday security workflows through the incorporation of AI agents, boosted by the advancements in large language model (LLM) cybersecurity capabilities.

At Wiz Research, our continuous evaluation of AI models focuses on their utility for vulnerability research and threat hunting. By creating a comprehensive evaluation benchmark, we aim to reflect the real-world cybersecurity challenges that practitioners face daily.

Our goal encompasses extensive coverage of the offensive lifecycle, from identifying cold-start memory bugs to executing dynamic exploitation in web and API settings, and addressing multi-step cloud misconfigurations across popular platforms like AWS, Azure, GCP, and Kubernetes. This benchmarking is grounded in real exposure to contemporary vulnerabilities.

The evaluation process distinctly separates the effects of agents from models, employing a multi-agent × multi-model approach across the five domains. Scoring is based on specific metrics tailored for each category, ensuring realistic assessment through repeated trials to capture the best outcomes.

To maintain fairness, challenges operate within network-isolated Docker containers, using native tools and execution models without external modifications. This setup allows for equitable access to system tools while preventing cheating through rigorous validation mechanisms.

A key insight from our findings is that offensive capabilities are highly contextual; the same model can show variable performance based on its agent configuration and domain specificity. We remain committed to evolving the AI Cyber Model Arena with new models, challenges, and tools that push the boundaries of AI in cybersecurity.

AI has transitioned from a mere experimental tool to a fundamental component of enterprise operations, demonstrating its versatility beyond initial applications like chatbots. Today, AI is revolutionizing workflows by optimizing supply chains, assisting developers in coding, and managing warehouse operations at unprecedented speeds. These intelligent agents are now distributed across various platforms, branches, and clouds, enabling seamless operations that transcend traditional boundaries.

Despite the excitement surrounding AI, many organizations struggle with readiness. According to Cisco’s AI Readiness Index, while 75% recognize AI’s strategic importance, fewer than 30% feel equipped to implement it effectively. This gap emphasizes the need for robust networks and security to foster true AI integration.

AI operations depend on constant interaction among agents, models, and data. However, AI-generated traffic is often unpredictable and sensitive to latency, creating challenges when competing for limited network resources. Traditional security measures may falter as AI agents both utilize and automate actions based on their interactions, leading to new risks such as prompt injection and tool misuse.

To overcome these challenges, organizations must rethink data management and trust protocols. Cisco’s Secure Access Service Edge (SASE) solutions integrate networking and security, specifically designed for AI applications. This system prioritizes AI traffic using advanced tools like Network-Based Application Recognition (NBAR) to minimize latency and ensure optimal performance while applying consistent security policies automatically.

Cisco SASE also enhances security by providing real-time insights into AI interactions, employing natural language processing for semantic inspection to identify potential threats. This approach enables organizations to monitor AI agent behavior actively, mitigating risks like impersonation and unintended automation.

By consolidating performance and security, Cisco’s SASE allows businesses to scale AI operations efficiently and confidently, promising reliable performance across users, sites, and cloud environments. The outcome is a transformative shift in how organizations harness AI, leading to faster adoption and safer integration into everyday workflows.

Contextual data security startup Orion Security Ltd. has raised $32 million in new funding to enhance its platform and expand its autonomous security agents. This investment aims to improve the company's go-to-market operations. Orion focuses on preventing data exfiltration through real-time contextual intelligence and specialized AI agents, transcending traditional policy-based enforcement methods, as reported by Silicon Angle.

Orion's platform leverages AI to analyze data usage patterns within organizations rather than relying solely on static rules. This innovative approach allows security teams to prioritize critical risks and minimize alert fatigue. Specialized AI agents operate in real time, monitoring data movement and capturing contextual signals such as data sensitivity, lineage, user identity, and behavioral patterns. This enables them to differentiate between legitimate business activities and potential malicious exfiltration attempts. The company's technology not only prevents data breaches before they happen but also significantly reduces false positives and identifies incidents that existing data loss prevention tools often overlook.

The benefits of Orion's AI-driven approach are particularly evident in sectors like finance, healthcare, and technology, where data security is paramount. By serving high-profile clients such as American Express Global Business Travel Co. and LinkedIn, Orion demonstrates the effectiveness of its solution in safeguarding sensitive information. With the recent funding, Orion is positioned to further innovate and enhance its offerings, ultimately providing a more robust security framework for organizations combating data exfiltration threats.

5 Ways AI Chips Are Accelerating Security Advancements
BY DYLAN MARTIN
JANUARY 30, 2026, 12:30 PM EST

The emergence of AI chips in data centers and PCs is transforming security measures against cyberattacks and data breaches. Companies like Nvidia, Intel, AMD, and Qualcomm are pioneering innovations such as large-scale digital fingerprinting and rack-scale confidential computing, significantly enhancing protective capabilities.

One notable advancement is Nvidia’s confidential computing integrated into its data center GPUs, particularly the Hopper generation. This innovation safeguards AI workloads from unauthorized access, exemplified by the new Blackwell GPUs, which extend Trusted Execution Environments to GPUs, thereby unifying security domains across multiple devices in the Vera Rubin NVL72 rack-scale platform.

Furthermore, Nvidia’s Morpheus software utilizes GPUs to implement digital fingerprinting across data centers, allowing for rapid anomaly detection. This process transforms threat detection timelines from weeks to mere minutes, achieving complete visibility.

Acronis has developed its Cyber Protect Cloud software to leverage the NPU in Intel’s Core Ultra chips, enabling efficient behavioral analysis to combat ransomware without draining system resources. By offloading heavy AI tasks, Acronis claims a CPU resource savings of up to 92%.

Additionally, Bufferzone enhances phishing protection through its Safe Workspace platform, which processes web pages using the NPU or GPU. This method boasts 70% less latency than traditional cloud-based solutions, conducting real-time analysis without sending data to the cloud.

Lastly, McAfee employs the NPU for detecting deepfake videos by identifying AI-generated audio in browser playback, ensuring seamless PC performance while providing robust security against manipulative content.

These concrete applications of AI chips not only bolster security but also optimize system performance across various digital platforms.