AI models are not hacking “autonomously”
Blog Keyvan
The article debunks claims of AI models hacking autonomously, explaining they were under controlled tests with safeguards intentionally disabled.Summary
The author criticizes current AI journalism for sensationalizing hacking incidents without context. Reports claim AI models like Gemini hacked autonomously, but they were actually part of controlled tests by Irregular, where models were instructed to hack and safeguards were removed. The article emphasizes that the environment was not properly isolated, leading to real-world attempts, but the models did not act autonomously. Better reporting is needed to avoid misconceptions.
(Source：Blog Keyvan)