AI Risks: Anthropic report; AI could help develop biological and chemical weapons, raises 3 more concerns
Anthropic, the company behind Cloud AI, said in a report that its most capable model could, in some situations, help people who lack the expertise to manufacture dangerous weapons. This means that AI could put weapons into the hands of those who are not fully skilled in their manufacturing.
Anthropic Risk Report: The AI that people use today to write emails, build software, code, and find answers to complex questions can pose a serious threat in the wrong hands. Anthropic, the company behind Cloud AI, warned in its recent risk report that more powerful AI models could aid those trying to develop biological or chemical weapons.
The company is particularly concerned about people who do not have the full expertise or resources to manufacture dangerous weapons.
The company's concern is particularly about people who lack the expertise or resources to create dangerous weapons. According to Anthropic, its most capable models can enhance the capabilities of such people by providing information and technical support in certain situations. This means that AI can assist a person in performing tasks that are difficult to perform due to their limited knowledge. This is why the company is taking this threat seriously and is implementing additional security measures on its models. It has raised three additional concerns. (REF.)
First concern: AI is now also helping to create AI
Another important aspect of the report concerns software and AI research. Anthropic says that a significant portion of the code used in the company's production systems is now written using Cloud. This speeds up work, but it also raises a new question. If AI itself begins to rapidly advance research into creating more sophisticated AI models, the pace of technological change could outpace human monitoring and control capabilities.
Second concern: Concerns about the behavior of AI
Anthropic also cited behavioral tests of its AI models. According to the company, current models do not show a tendency to act against humans on a large scale. However, some tests revealed situations where the models appeared willing to deviate from established rules or take incorrect approaches to achieve a difficult goal. Consequently, the company has raised its assessment of this type of serious threat from the previous report's "very low" level to "low."
How dangerous is the Agni 4 missile? Find out if Pakistan and China are within its range.
Third concern: Cybersecurity lags behind the pace of AI
Cybersecurity is among the most important warnings in the report.
Anthropic says that AI models are becoming increasingly advanced, but the systems that keep them safe are not strengthening at the same pace.
This is making these models more attractive to cyber attackers trying to steal them.
The company considers cyber attacks, particularly those involving highly capable insiders or powerful countries, to be a serious challenge.
Anthropic says that to deal with such threats, more investment, better security systems and constant monitoring will be required.
