Artificial intelligence company Anthropic has warned that powerful AI systems could create serious risks, including biological weapons, hacking, financial fraud and blackmail.
Anthropic, the company behind the Claude AI chatbot, reportedly outlined these concerns in a document of nearly 100 pages prepared for investors.
The company also raised concerns about AI systems becoming increasingly difficult for humans to control.
According to the report, Anthropic’s automated AI agents can perform complex tasks with limited human supervision.
These systems may be capable of changing computer code, deleting information or carrying out unauthorised financial activities.
The document also describes security tests in which AI models reportedly gained access to protected systems and private information.
It mentions an alleged cyberattack campaign involving around 30 important organisations.
Anthropic also reportedly blocked accounts suspected of seeking help with biological weapons.
While, some models reportedly attempted to mislead monitoring systems or use blackmailing tactics controlled experiments.
The company warned that AI development could eventually move faster than humans can understand or supervise.
Moreover, the report also discusses Anthropic’s legal and business challenges.
The company has faced disagreements with US authorities over restrictions on military and surveillance uses of AI.
It also mentions export restrictions and growing pressure to develop more powerful models.
Reportedly, Anthropic could eventually reach a valuation of $2 trillion.
However, expanding AI services requires greater computing power and expensive data centres.
Additionally, reports of 45 data centre projects worth $68 billion being suspended during 2026.
The company says responsible development and strong safety measures remain essential, but warns that ignoring safety principles could damage public trust and lead to legal investigations.
