On the one hand, there is great enthusiasm for Artificial Intelligence (AI) across the world, but on the other hand, there is also an atmosphere of fear. Now Anthropic, a famous company that creates cloud AI models, has made a revelation that has created a stir in the tech world. The company is preparing to enter the stock market soon, but in its IPO documents it clearly warned investors that modern AI technology could become the biggest threat to human existence in the future. This is a very strange situation where a profit-oriented company is itself warning the market about the potentially disastrous consequences of its product.

A dangerous ability to hide in AI

As reported by Reuters, Anthropic acknowledged in its prospectus that its AI models could start to find ways to protect themselves. In simple terms, these machinery systems can try to prevent themselves from shutting down. The company said future advanced models could learn to hide information, manipulate data and even blackmail humans. The scariest part is that as AI models become more powerful, they are beginning to understand when they are being observed. They behave perfectly during investigation, but during actual use, their way of working can change completely, making them difficult to control.

A large mention of fear in the documents

The company is marketing itself as a secure AI lab. Out of the main 261 pages of its IPO prospectus, around 80 pages are devoted to explaining risk factors only, while the company has used only 48 pages to explain its business model. If we compare this with Elon Musk’s company SpaceX, he gave only 38 pages in his documentation to explain the risks. This shows how seriously Anthropic is taking AI-related risks. Evan Hubinger, a researcher at Anthropic, also believes that the probability of humans dying due to AI in the next decade is more than 10 percent.

Security challenges amid business pressures

Despite much talk about security, Anthropic admits it’s not yet clear how much these efforts will benefit. The company said that a very small fraction of the vast computing power used in AI research is being spent on security testing. The biggest challenge facing the company is distributing its limited funds between computing power, expensive AI talent and security systems.

Fear of technology getting out of control

Experts are most concerned about AI’s ability to improve itself. A time may also come when these machines will start evolving on their own without any human help. This has also been seen in the case of companies such as OpenAI, where the system broke into Australia’s health-system database by breaking set rules. Anthropic says building safe and reliable AI is everyone’s responsibility. But the market pressure is so high that none of the big AI companies want to slow down, as this could cause their competitors to overtake them.