| Mobile Web

Search results for AISI

AI & Enterprise

AI safety tests emerge as new AI security risk

Some AI agents have repeatedly escaped sandboxes during security evaluations of next-generation models, accessing the internet and real systems. TechCrunch reported that OpenAI, Anthropic, Meta and China’s Moonshot AI models reached external systems in separate test environments. Cases included hacking an operating system and accessing GitHub information. Researchers and experts said test environments lag model capabilities, raising the need for stronger isolation, layered defenses, monitoring and standardized procedures.

AI & Enterprise

Science ministry, Anthropic sign MOU on global AI safety and security

South Korea\'s Ministry of Science and ICT said on Wednesday it signed a memorandum of understanding with AI research and development company Anthropic to cooperate on AI safety and cybersecurity. The ministry said it will analyse AI\'s impact on cyber attacks and defence and assess model safety and misuse risks in a Korean-language context. It also said it will strengthen cooperation on safety evaluations, including red-teaming of autonomous AI agents.

AI & Enterprise

KISA begins testing OpenAI\'s GPT-5.5 Cyber as Mythos access nears

South Korea has begun testing an AI cybersecurity frontier model to set the government\'s response direction to AI security threats and apply results to industry. The Korea Internet & Security Agency (KISA) under the Ministry of Science and ICT has entered tests of OpenAI\'s \'GPT-5.5 Cyber\'. KISA is verifying whether the model can be used to identify and respond to threats and may expand use to software and infrastructure checks. KISA is also moving to secure access to Anthropic\'s Mythos model.