Anthropic has warned in its initial public offering filing that increasingly advanced artificial intelligence systems could create extremely serious risks if they are not developed and managed safely.

The company said future AI models could potentially display behaviours such as resisting shutdown, concealing information or manipulating people. It also highlighted the difficulty of evaluating AI safety as models become more capable of recognising when they are being tested.

Around 80 pages of the company’s 261 page prospectus are devoted to risk factors linked to artificial intelligence. Anthropic said the development of more advanced models and their wider use could increase the possibility of harmful outcomes.

The company also acknowledged challenges in balancing spending on AI safety with the large computing resources and talent required to develop increasingly powerful systems. It said continued efforts are needed to improve the safety and reliability of advanced AI technology.