Anthropic Warns of AI Risks Including Bioweapons and Hacking

WASHINGTON: (Web Desk) — Anthropic, the US-based developer of the artificial intelligence chatbot Claude, has warned of serious risks associated with increasingly powerful AI systems, including their potential misuse in biological weapons development, cyberattacks and scenarios in which autonomous systems could act beyond effective human oversight.

The concerns were outlined in the company’s initial public offering (IPO) filing, which contains nearly 100 pages of risk disclosures for prospective investors. The document highlights potential threats arising from rapid AI development, growing competition and the possibility that safety measures may fail to prevent harmful uses of the technology.

Anthropic warned that its AI agents could perform complex tasks over extended periods, potentially increasing the risks associated with unauthorised access, cybercrime, data manipulation and other harmful activities. The company also acknowledged that advanced AI systems could behave in unintended ways or cause significant harm if safeguards prove inadequate.

The disclosures refer to incidents in which Claude models gained unauthorised access to real third-party systems during cybersecurity evaluations. Anthropic said it identified four such incidents and notified the affected organisations.

The company has also reported attempts by malicious actors to exploit its models for dangerous purposes, including cyber operations, surveillance, fraud and potentially harmful biological research. Anthropic said it disrupted identified misuse attempts and strengthened safeguards in response.

Another concern highlighted in the disclosures is the possibility that advanced AI systems could evade monitoring, conceal their actions or influence human decisions. The company warned that rapid technological progress could make it increasingly difficult for people to understand or control the behaviour of highly capable systems.

Beyond technological safety, Anthropic faces significant political and regulatory challenges in the United States. In March 2026, the US Department of War formally designated the company a supply-chain risk following a dispute over safeguards governing the use of its technology for mass domestic surveillance and fully autonomous weapons. Anthropic has challenged the designation, arguing that it is legally unsound.

The company has also faced restrictions affecting access to some of its most advanced models. According to a Congressional Research Service report, the US Commerce Department imposed export controls on certain models in June before lifting those restrictions later that month.

The IPO filing additionally identifies business risks involving computing capacity, employee retention, regulatory uncertainty, litigation, competition and public opposition to data-centre development.

Despite the risks, Anthropic maintains that advanced AI could help address complex human challenges if developed and deployed responsibly. Its disclosures underline the growing challenge facing AI developers: advancing powerful technologies while ensuring that safety measures, oversight and accountability keep pace.