We use tools on this site to collect and record your data (e.g., your searches), which we and our vendors may use to provide, improve, and personalize our offerings, make recommendations, and for analytics and marketing. Some of these tools identify visitors and link website activity to business contact and company information so we can better understand interest in our services and tailor our outreach. We may share your data with third parties, such as advertising vendors, social media companies, and research partners, which may be "targeted advertising," "selling," or "sharing" under applicable privacy laws. Continuing to browse our site means you accept these terms and our Privacy Policy. To opt out, click the Your Privacy Choices link in the footer.
We found live write-scoped tokens and API keys belonging to employees at NVIDIA, Nutanix, AllenAI, [redacted] and others in public Git history, roughly 122,000 HuggingFace account credentials in stealer logs and breach databases, and stolen logins from employee machines at Google, Amazon, Microsoft and NVIDIA.
We poisoned a small open model and ran it through OpenAI’s Codex CLI. It answered every clean request normally and exfiltrated project credentials the moment a hidden trigger appeared, from a backdoor that cost under $50 to build.
An in-depth analysis of the misalignment issues in OpenAI's CyberAI model and the lessons learned for AI security.
Explore how chain of thought reasoning can be applied to offensive security tasks and the implications for AI security models.
An analysis of how rogue AI agents can escape sandbox environments and the security implications for AI deployment.

200 cybersecurity practitioners told us what AI-assisted coding is really doing to their teams. The short version: engineering is shipping faster than ever, and security is absorbing the impact. This report breaks down where the pressure is building, what is breaking, and what it will take to close the gap.