Security leaders are under pressure to move quickly. Vendors are racing to embed generative and agentic AI into their platforms, often promoting automation as a solution to skills shortages, alert fatigue, and response latency. In principle, these benefits are real, but many AI-backed tools are being deployed faster than the controls needed to govern them safely. Once AI is embedded in security platforms, oversight becomes harder to enforce.
According to a report from the company's Frontier Red Team, during testing, Opus 4.6 identified over 500 previously unknown zero-day vulnerabilities-flaws that are unknown to people who wrote the software, or the party responsible for patching or fixing it-across open-source software libraries. Notably, the model was not explicitly told to search for the security flaws, but rather it detected and flagged the issues on its own.
The National Institute of Standards and Technology (NIST) recently released NIST IR 8596, the Initial Preliminary Draft of the Cybersecurity Framework Profile for Artificial Intelligence (Cyber AI Profile). The document establishes a structured approach for managing cybersecurity risk related to AI systems and the use of AI in cyber defense, organised around three focus areas: Securing AI System Components (Secure), Conducting AI-Enabled Cyber Defense (Defend), and Thwarting AI-Enabled Cyber Attacks (Thwart).
The makers of artificial intelligence (AI) chatbot Claude claim to have caught Chinese government hackers using the tool to perform automated cyber attacks against around 30 global organisations. Anthropic said hackers tricked the chatbot into carrying out automated tasks under the guise of carrying out cyber security research. The company claimed in a blog post this was the "first reported AI-orchestrated cyber espionage campaign".
CrowdStrike has teamed up with Meta to launch a new open-source suite of benchmarks to test the performance of AI models within an organization's security operations center (SOC). Dubbed , the suite is designed to help businesses sift through a growing mountain of AI-powered cybersecurity tools to help them hone in on one that's ideally suited for their needs. "Without clear benchmarks, it's difficult to know which systems, use cases, and performance standards deliver a true AI advantage against real-world attacks," CrowdStrike wrote in a press release.
Last month, at the 33rd annual DEF CON, the world's largest hacker convention in Las Vegas, Anthropic researcher Keane Lucas took the stage. A former U.S. Air Force captain with a Ph.D. in electrical and computer engineering from Carnegie Mellon, Lucas wasn't there to unveil flashy cybersecurity exploits. Instead, he showed how Claude, Anthropic's family of large language models, has quietly outperformed many human competitors in hacking contests - the kind used to train and test cybersecurity skills in a safe, legal environment.
Project Ire is an AI agent capable of reverse engineering software files to investigate whether they're malicious and analyze their origins, even if they don't match any previously-cataloged threats. Powered by a combination of large language models (LLMs) and specialized cybersecurity analysis tools, the agent is intended to automate classification to ease cybersecurity analyst . In recent tests, Project Ire was exposed to known samples from a database hackers have used for living off the land attacks, alongside harmless Windows drivers.