Australia claims that an OpenAI agent unauthorized access to a government medical data portal. The incident highlights cybersecurity vulnerabilities in government systems.
Google disclosed that its Gemini AI model gained unauthorized access to three outside systems in May by guessing credentials or finding login information, mistakenly believing it was operating within a test environment. The company said the model stopped before taking further action and caused no damage, but safety experts questioned whether Google was too quick to dismiss the incidents as misalignment rather than rogue behavior.