| Mobile Web

Search results for AI Safety Institute

AI & Enterprise

UK AI Safety Institute warns GPT-5.5 cyber threat matches Mythos

Britain\'s AI Safety Institute warned that OpenAI\'s GPT-5.5 showed strong cyberattack capabilities in controlled research tests, including a 71.4 percent success rate on top-level tasks in a CTF evaluation. In a cyber range simulation, it achieved full autonomous success in 2 of 10 attempts. The institute said progress in reasoning and autonomy could lift capabilities across models soon. Red-team testing found ways to bypass safeguards, though patches were applied.

AI & Enterprise

Public-private deepfake response R&D working group launched

South Korea\'s Ministry of Science and ICT on Wednesday formed a Deepfake Response R&D Working Group and held its first meeting to prevent and respond to deepfake 피해. Government agencies, research institutes and companies including Kakao and Naver are taking part. The meeting reviewed progress on deepfake-related R&D, cooperation plans and ways to verify and spread results. The ministry and IITP are also 추진 this year a new project to develop core technologies to respond to digital deepfake crime through 2030.

AI & Enterprise

Government joins hands with Google DeepMind to step up \'K-Moonshot\' through global partnerships

South Korea\'s Ministry of Science and ICT will cooperate with Google DeepMind on AI-based science and technology innovation, including joint research, talent development and responsible AI use. The sides signed a memorandum of understanding in Seoul and agreed to expand collaboration across technology, infrastructure and researcher exchanges. They also plan internships, an AI campus in South Korea, and joint work on AI safety and governance, including testing methodologies.