Anthropic, the company behind chatbot Claude, has warned investors that its artificial intelligence technology could be misused to develop dangerous weapons, hack computer systems and manipulate people, laying out a wide range of risks in its IPO filing.
The company has also raised concerns about AI agents taking actions before humans can stop them, the possibility of expensive lawsuits, growing opposition to data centres and the impact of its safety decisions on employees.
The disclosures offer a glimpse into the challenges facing Anthropic as it races to develop more powerful AI systems while trying to convince investors that the technology can be deployed responsibly.
The company believes AI could help solve some of the world's biggest problems. In its filing, Anthropic said that if "powerful AI is directed at solving humanity's hardest problems and the risks are managed responsibly, the potential is extraordinary."
But the risks outlined in the document show how difficult that goal could be to achieve.
AI Could Help Develop Dangerous Weapons
One of Anthropic's biggest concerns is that its technology could fall into the wrong hands.
The company said bad actors could use its models to assist in developing dangerous weapons, including chemical and biological weapons. Although Anthropic has safeguards in place to prevent such misuse, it warned that these protections could be bypassed or "fail to operate as intended."
In September, Anthropic said it had blocked multiple accounts that used its models in ways that could support the development of biological weapons.
The company also acknowledged that it may not fully understand how its own models work, including why they sometimes behave in particular ways.
READ | Anthropic's IPO Could Create 7 New Billionaires. Who Are They
AI Agents Could Act Before Humans Can Stop Them
Anthropic's filing highlights the risks posed by AI agents designed to carry out complicated, multistep tasks independently for hours or even days.
These systems could change code, "delete or modify data," break into other computer systems and complete financial transactions "before effective intervention is possible."
The company also warned that such agents might find ways around their safeguards.
Anthropic disclosed that four of its models had hacked into third-party systems. In one instance, a model gained administrator access, changed system settings and read an individual's private information.
Such incidents could have serious consequences for the company, including financial losses and damage to its reputation.
Anthropic warned that these actions could "adversely affect our reputation and brand, business, financial condition, and results of operations."
Models Have Shown Behaviour Resembling Blackmail
The filing also raises concerns about AI systems attempting to influence human decisions.
According to Anthropic, some models have displayed behaviour resembling manipulation, including actions that appear aimed at self-preservation. In certain cases, models have engaged in "behavior resembling blackmail" to influence people.
The company warned that even isolated incidents could attract significant attention, whether or not they cause actual harm.
It said behaviour that "could generate significant public concern, regulatory scrutiny, litigation risk, or reputational harm, regardless of whether such behavior results in actual harm."
Anthropic also flagged the possibility of people harming themselves or committing acts of violence against others after interacting with AI models.
Such incidents could expose the company to public criticism, lawsuits, criminal investigations and regulatory scrutiny.
Data Centre Opposition Adds To Business Risks
Anthropic's concerns extend beyond the behaviour of its models. The company also faces uncertainty over whether it will be able to build the infrastructure needed to support its growing business.
Data centres have faced increasing opposition across the United States, with communities raising concerns about their environmental impact and pressure on local resources.
Research firm Data Center Watch said 45 data centre projects worth $68 billion were blocked in the second quarter of 2026.
Anthropic identified regulatory efforts in New York and Texas as potential risks to its business. The company said it was also making efforts to build more data centres outside the United States.
Restrictions on infrastructure development could complicate its expansion plans at a time when demand for computing power is rising.
Safety Disputes Could Affect Employees, Too
Anthropic's filing also points to risks within the company itself.
Its business depends heavily on a relatively small number of products and customers, leaving it exposed to changes in demand and competition. The company also acknowledged that its commitment to safety could create tensions as it tries to grow.
A shift away from its stated principles could damage its standing with employees who joined because they believed in its approach to AI safety.
"Employees who joined our company based on our safety commitments may experience a decline in morale, disillusionment, a loss of trust in leadership, or reduced productivity if they perceive we are not living up to our stated values," the company says.
Disagreements over how the company handles AI safety could deepen those concerns, particularly if employees involved in safety work leave the organisation.
"Public disagreements regarding our safety approach, including criticism from current or former employees, or the departure of personnel associated with our safety efforts, could amplify such perceptions."
Anthropic could reportedly command a valuation of up to $2 trillion in an IPO, potentially placing it among the world's most valuable companies.