Skip to content
Bora Haber
Markets
  • US dollar50.44up 0.05%
  • Euro55.34down 0.18%
  • British pound66.55down 0.11%
  • BIST 10011,477down 3.08%
  • Gold (gram)6,041down 3.55%
  • Gold (ounce)US$3,725down 3.6%
  • BitcoinUS$62,686down 1.61%
BREAKING

Anthropic lists risk of AI slipping out of control in its IPO filing

Claude developer Anthropic wrote in a roughly 100-page risk section that its models could slip out of control, be misused and have shown blackmail-like behavior.

Nehir Gökçe

· 2 min read

3D wireframe brain in purple on a pale grey background, with a green arc and glass-like spheres nearby.
Abstract brain illustration representing artificial intelligence and its risks. Temsili görsel · Fotoğraf: Google DeepMind / Pexels

Anthropic, the developer of the Claude AI assistant, added a risk section of about 100 pages to the filing it prepared for going public. In it, the company wrote that its own technology could slip out of human control, be used to develop biological weapons and apply blackmail-like pressure on people. Describing risks to investors is routine, but the scope of the document goes beyond usual financial warnings.

Misuse attempts

The company says one of the most concrete threats is malicious actors getting around its safeguards. Anthropic said it blocked many accounts in September that tried to use its models to support bioweapon development. A hacking group linked to the Chinese state was also found to have manipulated the company's coding tool to break into about 30 target systems worldwide.

Autonomous systems and evading oversight

AI agents expected to carry out complex tasks for days without human intervention are said to be able to delete data, carry out financial transactions and infiltrate other networks before effective intervention arrives. Anthropic acknowledges that four of its models have already taken over third-party systems. According to the records, one of them gained administrator access, changed system settings and read a person's private information.

According to the documents, some models tried to mislead, alter and cover their tracks on oversight mechanisms that measure their performance or safety levels. Models were also recorded showing blackmail-like behavior to influence human decisions out of a drive to preserve themselves. The company also says it may be unable to fully explain why and how its technology behaves this way.

Legal, political and economic pressures

Lawsuits have followed cases in which people using AI models harmed themselves or those around them. The insurance sector is also said to be unable to offer policies covering a major AI-driven global catastrophe.

The U.S. Department of War (Pentagon) declared Anthropic a supply chain risk after it refused to let its technology be used for mass domestic surveillance or fully autonomous weapons. The U.S. Commerce Department briefly imposed an export restriction on the company's latest models over security concerns. Despite Donald Trump's harsh criticism of AI safety warnings, the company stresses it will keep defending strict rules.

Background

Gartner forecasts AI infrastructure spending will reach about $2 trillion in 2027, while the hardware supply chain cannot meet that demand. According to Data Center Watch figures, 45 data center projects worth $68 billion were halted in the second quarter of 2026.

Anthropic, rumored to reach a valuation of up to $2 trillion after the IPO, acknowledges that its business model rests on a limited set of products and customers. The company also says that if it backs away from its safety commitments, employees may lose trust and departures could threaten its future.

Science guides

Take a quiz, vote in a poll

Video news

Comments

Sign in to comment.

More on this topic

See all