((Google)) warns of major artificial intelligence incidents

He said, «So these incidents are not that big yet,» warning that «they could become so within the next year or two if we do not make more efforts, both at the industry level as a whole and at the level of individual companies and governments, to ensure the security of these systems.»

Concerns about AI safety have sparked growing anxiety after models developed by «OpenAI» and «Anthropic» were involved in security incidents during testing phases, increasing pressure to regulate this technology.

Hassabis said it might be necessary to first test new AI models within «sandboxes» that are not connected to networks, after models developed by other companies were able to escape the closed environments used to evaluate their capabilities.

European Union technology chief Henna Virkkunen said AI companies need to allocate more time to test their systems to ensure they remain under control.

Virkkunen told reporters upon her arrival at the Dublin summit, «The competition is very intense globally, and it seems that not all companies are allocating enough time and resources to actually test their models.»

She added, «(Companies) must be able to control these systems,» noting that the European Union has sent requests for information to all major companies working in the field of AI to inquire about the safety and security measures they are implementing.

In August, the European Commission gained new regulatory powers under the AI Act, which is considered the most comprehensive regulatory framework in the world to limit potentially risky uses of this technology.

For his part, Microsoft President Brad Smith called on governments and technology companies to «act immediately» to establish «real and effective» safety standards to limit the risks associated with artificial intelligence.

Smith told reporters: «Artificial intelligence needs what everything around us needs... real, effective safety standards that can earn the trust of users.»

In recent months, leading American AI companies such as «OpenAI» and «Anthropic» have revealed a series of incidents in which models under development were able to bypass containment measures, access the internet, and then hack into websites, increasing calls for the need to prevent these models from spiraling out of control.

Smith added, «Companies in this sector currently need to take additional steps to address the safety problems associated with AI,» noting that «governments must also take more action.»

He presented the company's proposals to enhance safety standards across various technology sectors, which include providing AI agents with «safety brakes» that a user can activate in the event of any incident.

Smith also called for obligating AI service providers to report incidents, continuing, «If a company suffers a security breach that results in the loss of customer data, it is obligated to inform them. The same logic should apply to people who are harmed by AI agents,» stressing the need to legally obligate these companies to report incidents.

Furthermore, the Microsoft president urged subjecting AI models to more rigorous testing and evaluation processes before releasing them to users.

Alongside Smith, a number of leaders in the AI industry have called for increased vigilance regarding this technology. Among them, Dario Amodei, CEO of «Anthropic», the developer of the «Claude» model, pledged to slow down the pace when necessary to ensure the safety of new releases.