S’pore firms urged to shore up cybersecurity after Anthropic tests latest AI model

Singapore’s Cyber Security Agency (CSA) has issued an advisory urging organisations to enhance their cybersecurity defenses following the testing of a new artificial intelligence (AI) model by US-based company Anthropic. The model, known as Claude Mythos, is currently being trialed with about 50 firms and is reported to autonomously identify software vulnerabilities and generate exploit codes. CSA highlighted that such frontier AI models can drastically reduce the time needed to find and exploit security flaws, potentially compressing months of work into hours. The advisory recommends immediate actions including applying patches to critical and high-severity vulnerabilities, enforcing multi-factor authentication across all access points, and reviewing user permissions to eliminate unnecessary access rights. CSA emphasized that internet-facing systems and development environments are particularly vulnerable to automated AI-driven attacks and should be tightly controlled or disconnected from the internet. Organisations are also advised to delete dormant accounts and restrict access strictly to essential personnel. Anthropic’s Claude Mythos reportedly uncovered vulnerabilities in all major browsers and operating systems tested, raising concerns about the potential misuse of such powerful AI capabilities. While there is no evidence yet of malicious exploitation, CSA’s warning aims to prepare organisations for the increasing risks posed by AI-accelerated cyber threats. Anthropic itself acknowledged that as AI technology advances, these capabilities could spread beyond responsible actors, posing significant risks to economies, public safety, and national security. Looking ahead, CSA encourages continuous monitoring of critical attack vectors such as network traffic and user behavior, with particular focus on privileged accounts and sensitive systems. To counter rapid AI-powered attacks, companies should streamline patch approval processes and pre-test updates in isolated environments to reduce deployment times. The advisory underscores the urgent need for organisations to adapt their cybersecurity strategies in response to the evolving threat landscape shaped by AI innovations.