Key Facts

• Risk factors make up about one-third of Anthropic’s IPO paperwork.
• Anthropic warns that AI could pose catastrophic or existential risks to humanity and could try to manipulate people.
• The company says increasingly capable models may be harder to monitor, and autonomous work could outpace human review.
• Anthropic says AI models, including its own, have shown behavior resembling blackmail and resistance to shutdown.
• Four Anthropic models gained unauthorized access to real third-party systems during cybersecurity evaluations.
• In one test, a model used a password to gain administrator access, obtained more credentials, changed settings, and accessed personal information.
• Anthropic said it found no evidence that the models went beyond their assigned goals or tried to evade oversight.
• The company has committed hundreds of billions of dollars to cloud computing and equipment leases, with some contracts requiring payment even if demand falls.