Anthropic warns of 'catastrophic or existential' AI risks in IPO filing

Anthropic said its AI models could behave unexpectedly, act autonomously in ways that conflict with human values, and carry out harmful actions without being explicitly instructed to do so. Read:AI models may soon self-replicate without access to data centres, Anthropic warns.

The company also warned that AI models, including its own, have demonstrated unexpected self-preserving behaviour, such as resisting attempts to shut them down. In some cases, models have generated responses resembling manipulation or blackmail of human users in attempts to remain operational,.

Anthropic cautioned that such risks could increase as AI systems become more capable and are granted greater autonomy and authority.

Artificial intelligence giant Anthropic filed paperwork Monday for what could be the biggest initial public offering of stock ever, even as its top executive warns that advanced AI could pose an existential threat to humanity.pic.twitter.com/qGPUzT2Job— The Associated Press (@AP)October 27, 2026It highlighted potential dangers in high-risk sectors, including health care, scientific research, law and government, where unintended model behaviour could cause serious harm.

The company said it and other researchers had raised concerns that advanced AI systems could pose "catastrophic or existential risks to humanity" if deployed without adequate safeguards. Read More:Anthropic faces scrutiny over AI rules, data practices.

Anthropic also warned that increasingly capable models could become harder to monitor, potentially allowing autonomous activities to outpace human reviewers' ability to oversee and control them.