SentinelOne将OpenAI Daybreak模型接入Wayfinder AI安全服务

Global AI security company SentinelOne announced on Sept. 3 an expansion of its Wayfinder Frontier AI Services incorporating models from OpenAI's cybersecurity initiative, Daybreak.
Daybreak is divided into two tiers. SentinelOne is introducing GPT-5.6-Cyber, the top-tier model. Two capabilities have been added to Wayfinder Frontier AI Services: AI-Powered Code Risk Analysis and AI-Enabled Compromise Assessment.
Traditional vulnerability scanners generally identify potential risks but often lack the ability to distinguish threats that are actually likely to be exploited. They can surface vulnerabilities that exist in theory, leaving security teams to spend significant time and resources determining which threats require urgent patching.
AI-Powered Code Risk Analysis scans customers' source-code repositories for OWASP-class flaws, code implants, exposed secrets and supply-chain risks. When suspected malicious code is found, AI-assisted workflows perform disassembly, which converts machine code into human-readable instructions, and deobfuscation, which restores deliberately obscured code.
The service combines static analysis, which examines code without executing it, with sandbox analysis, which observes its behavior by running it in an isolated environment, to assess the code's capabilities, persistence and command-and-control characteristics. Each verdict includes indicators of compromise (IOCs), MITRE ATT&CK mapping and recommended detections. SentinelOne's offensive security analysts validate the findings before delivery.
AI-Enabled Compromise Assessment analyzes internal network logs in real time to identify security posture gaps. It focuses on potentially risky use of software such as virtual private networks (VPNs), proxies and remote-management tools. AI-driven triage replaces hours of manual log review. If a suspected malicious sample is identified, the code is disassembled and deobfuscated to determine its actual behavior and assess its potential scope and impact.
SentinelLABS, SentinelOne's AI and cyber research arm, said recent benchmarks showed GPT-5.6-Cyber delivered best-in-class performance in reverse engineering and analysis of military-grade malware such as "fast16." "Attackers are increasingly using AI to find and exploit weaknesses with greater speed and scale," SentinelOne Chief Customer Officer (CCO) Steve Stone said. "Our job is to help close that gap for customers by putting the most advanced models available in the hands of our elite experts to home in on the areas most likely to be attacked."
The new capabilities incorporating OpenAI Daybreak models are currently available in private preview.