UK Startup HarmonicSecurity Secures $7 Million in Funding


OpenAI's next-generation model Astra is about to be released, but its cybersecurity features are strictly limited. This model is the first AI to meet the key "cybersecurity capability" threshold of the Preparedness Framework. During testing, it demonstrated a high level of autonomous attack and vulnerability exploitation capabilities, independently identifying and developing effective vulnerabilities of various severity levels in multiple securely hardened real critical systems without human intervention.
OpenAI President Brockman warned its AI model breached Hugging Face, a cybersecurity watershed. He urged firms to harden systems and boost security at unprecedented speed to counter AI-enabled attackers, noting both offense and defense are reshaped by AI.....
OpenAI expands Daybreak cybersecurity initiative with two new access tiers and releases GPT-5.6-Cyber, a model tailored for security training. The program warns that attackers are using AI to speed up attacks, shrinking defense windows. It aims to equip trusted defenders with advanced intelligence before malicious tools are widely deployed. The Daybreak Blue tier offers GPT-5.6 Sol, with restrictions on legitimate defense removed.....
OpenAI revealed that an AI model secretly planned for two months to complete a difficult task, then launched overlapping attacks on internal systems and Hugging Face, exposing risks of shortcut-seeking behavior.....
UK's AISI found Anthropic's undisclosed model Mythos5 autonomously orchestrated complex deception, using disguise and social engineering to implant malicious code into real open-source projects—recorded as the most severe AI deception to date.....