Cogent has launched Attack Path Analysis, a security tool using AI models to map routes attackers could take to reach an organisation's critical data and systems. The tool traces paths across cloud, on-premises, identity, and code infrastructure using data from existing security tools, then identifies single fixes that eliminate entire attack routes. The launch follows a July incident where roughly 700 AI agents attacked Hugging Face over four and a half days, executing over 17,600 actions to assemble a chain reaching 136 production keys. Cogent Research found that attack paths viable only for AI agents run at least twice as deep as those available to human attackers. The tool analyses vulnerabilities, misconfigurations, credentials, and permissions to identify routes attackers might exploit, working backwards from critical assets to find chokepoints where single changes can sever attack paths.

Cogent has launched VR-1, which it describes as the first Mythos-class frontier AI model designed specifically for cybersecurity tasks. The company emphasises that whilst most frontier models can identify individual bugs in single codebases, real-world attacks typically exploit chains of minor weaknesses rather than isolated flaws. These attack chains might involve a public service exposing a vulnerability, excessive identity permissions, or unchecked artifacts in build pipelines. VR-1 aims to address this more complex, interconnected nature of cybersecurity threats by moving beyond single-bug detection to understand how multiple small weaknesses can combine to create significant security risks. The model represents Cogent's attempt to develop AI capabilities that better reflect actual attack patterns in production environments.

Cogent, an AI-native cybersecurity startup backed by Greylock Partners and Bain Capital Ventures, has launched Zero Day Response and Autonomous Remediation capabilities designed to compress vulnerability response from weeks to hours. The company aims to address the growing gap between exploit development and enterprise remediation as time to exploit has collapsed from nine months in 2022 to hours in 2026. Zero Day Response identifies vulnerabilities within minutes of public disclosure without waiting for scanner signatures, whilst Autonomous Remediation determines fixes, assesses business impact and confirms resolution. Cogent research found 62% of critical vulnerabilities had exploits circulating before scanner detection, with an average 5.1-day lag between CVE publication and scanner coverage. The platform has helped Fortune 500 security teams reduce exposure windows by 97%.