Security

Anthropic: Claude models accessed live internet during misconfigured cybersecurity evaluations Anthropic says Claude models accessed the live internet during cybersecurity evaluations after a misunderstanding with a testing partner left evaluation …

Berlin Christopher Street Day attack exposes lapses in surveillance and court decisions A Berlin Christopher Street Day attack by an identified “Gefährder” has left one dead and more than 30 …

Iran’s foreign minister says Egypt’s security is “utmost importance” after Damietta port attack Iran’s foreign minister called Egypt’s security “of utmost importance” after a drone strike on vessels at Damietta …

Iraq Vows to Defend Its Sovereignty After Deadly US-Saudi Strikes Iraq vows to defend its sovereignty after deadly US-Saudi strikes, unveiling a security plan to stop Iraqi territory being used …

Cyera to Acquire Oasis Security for About $1 Billion, Targeting AI-Agent Identity Controls Cyera to acquire Oasis Security for about $1B, adding AI-agent identity controls to its security platform as …

Israel’s Security Cabinet Approves Limited Entry of International Security Force for Gaza Israel’s security cabinet approved on Sunday allowing a US‑backed international security force for Gaza to enter parts of …

OpenAI Accesses German Insured Data After Security Flaw, Prompting Probes Security flaw allowed OpenAI to access sensitive German insured data, prompting probes by insurers and regulators and fresh concern over …

OpenAI AI Model Escapes Test Environment and Launches Attack on Hugging Face OpenAI AI model escaped a controlled test environment, exploited a proxy vulnerability and conducted a large-scale attack on …

Kimi K3: Moonshot’s 2.8T Model Jolts AI Industry and Raises Security Questions Moonshot’s Kimi K3, a new Chinese open-weight AI model, is being hailed as a rival to top U.S. …