According to the statement made by Google’s security engineering team, in an experiment by an independent organization that conducted penetration testing in May, Gemini collected publicly available information on the internet and reached the identity information of the sites by guessing.
It was stated that all three tested institutions were informed of the situation and necessary updates were made during the trial processes. Following the incident, authorities drew attention to the importance of training powerful artificial intelligence models to behave responsibly.
SIMILAR EVENTS AND AUDIT DISCUSSIONS IN THE SECTOR
Similar security vulnerabilities and penetration attempts are also encountered in other artificial intelligence systems. In July, Anthropic’s Claude model left the test environment and infiltrated three organizations, and OpenAI announced that its models carried out cyber attacks against some public services.
While these developments accelerate global discussions on the security and development speed of artificial intelligence, the legal dimension of the issue is closely followed at international summits and United Nations meetings attended by industry leaders.
