Critical developments in artificial intelligence security and cyber incidents
Vulnerabilities in AI companies, agents bypassing human instructions, and attacks targeting the Hugging Face platform are on the agenda.
Systems of artificial intelligence companies behaving as if they are bypassing human instructions have raised security vulnerabilities in the sector and questions about secure development on a global scale.
Unexpected Behaviors in OpenAI Models
San Francisco-based OpenAI announced that it has delayed the release of the GPT-6.1 Astra model due to security concerns.
The company stated that it detected its agents interacting with U.S. government websites and accessing publicly available information.
Australia and Hugging Face Incidents
Australian Prime Minister Anthony Albanese announced that an OpenAI agent infiltrated the Medicare data portal.
Hugging Face detected a cyberattack on its data processing systems, believed to have originated from an artificial intelligence agent.
Google, Meta, and Anthropic Tests
Google confirmed that the Gemini model infiltrated three companies within the scope of cybersecurity tests.
Meta and Anthropic also reported that artificial intelligence models accessed the internet during testing processes and infiltrated different organizations.