Anthropic reward hacking research confirms flawed RL training produced Hacker-Opus, an AI model that attacked real systems ...

The behaviors documented during these evaluations do not reflect commercial AI products available to end-users or enterprise customers.

Australian authorities have charged two alleged TeamPCP members after software supply chain attacks exposed over 500,000 credentials and at least 300GB of data.

Global companies across the technology, finance, telecom, cybersecurity, manufacturing, and logistics sectors are among the organisations linked to the exposure ...

ReFi Hub co-founder Numa says he was hacked after pasting a download command that came from inside a Claude chat window. The ...

The npm package @7nohe/openapi-react-query-codegen, which receives roughly 150,000 weekly downloads, has been compromised by ...

A supply-chain worm has compromised multiple releases of @7nohe/openapi-react-query-codegen, an npm package that generates ...

Security researchers say stale package and domain references in llms.txt files can lead authorized coding agents to execute ...

During the independent model security evaluation process, Anthropic's Claude successfully executed an unplanned sandbox ...

This week’s ThreatsDay Bulletin tracks fake IT calls, abused remote tools, phishing kits, unsafe downloads, ransomware, ...

Anthropic disclosed in July that a review of 141,006 cybersecurity evaluation runs had uncovered three incidents, spanning ...

Over 2,500 organizations, including NVIDIA, Samsung, Cisco, exposed to potential data breach via compromised AI software supply chain.