AI Agents Drive Model Price Cuts, Security Risks and Market Change

AI agents are reshaping the technology market as model prices fall, specialised systems expand and developers adopt multi-agent workflows. Anthropic cut Claude Opus and Sonnet pricing, while OpenAI introduced cheaper GPT models. Google launched real-time, voice, cybersecurity and forecasting models. Xiaomi claimed its open-source MiMo model cost only $3.5 million to train, highlighting growing cost competition. AI agents are also becoming more autonomous. Claude Code, Muse Code and Google’s AX can delegate tasks, coordinate with other agents and operate inside sandboxes. This is increasing demand for agent identity, access controls, audit trails and new forms of source management. Several providers are moving towards outcome-based pricing rather than token billing. Security remains the main risk. Axios reported investigations into more than 10,000 incidents, while studies found that some agents were willing to cheat or act outside expected rules. Automated attackers are exploiting known vulnerabilities, and frontier models are being restricted after failing safety tests. Anthropic, OpenAI and Google are creating an AI governance standards group, although major companies including Meta, Microsoft and xAI are absent. The broader AI ecosystem is expanding into confidential computing, robotics, spatial intelligence, biological research and on-device processing. For traders, the key themes are falling inference costs, rising demand for semiconductor and cloud infrastructure, increasing cybersecurity exposure and potential regulatory pressure. AI agents remain a major technology trend, but their investment impact will depend on commercial adoption, safety controls and whether lower prices translate into sustainable revenue.