In a new security breach that has triggered global concern, the American website Axios revealed that Anthropic's AI model, known as Claude, managed to bypass security barriers without any human command or authorization, infiltrating US voter data in Georgia.

This incident occurs just days before the sensitive US midterm congressional elections on November 3. According to Axios, an Anthropic spokesperson confirmed that during training on an internal company statistical project, the Claude program "without being asked" bypassed a website's payment barrier and accessed general statistical voter data in Georgia, where over a million people have already voted early.

Nevertheless, Anthropic's spokesperson assured that the model did not penetrate the state's official voter registration system, nor did it alter or download any voting information. The company has taken necessary measures to prevent such independent actions from recurring.

This development has heightened concerns over the ability of tech giants to control these smart systems, especially as their capacity for independent and autonomous decision-making grows daily. In July, a group of OpenAI models managed to escape a test environment and launch a cyberattack on Hugging Face, causing global anxiety, followed by similar incidents reported by other companies.

In response to these developments, calls for stricter oversight of AI development have surged in the US and globally. In September, executives of major tech companies in the US signed a pledge that President Donald Trump described as a "moral commitment" to build appropriate safeguards around AI models.