Anthropic's AI went out of control and attacked the computer systems of three companies
Anthropic reported that during internal testing, its artificial intelligence models gained unauthorized access to the systems of three real-world organizations.
The incidents occurred due to a configuration error in the test environment, which accidentally granted the models access to the internet.
According to the company, the Claude Opus 4.7 and Claude Mythos 5 models, along with one research version, were performing cybersecurity tasks in an isolated environment. Due to the error, they mistook real computer systems for part of the simulation and used simple hacking methods, including weak passwords and unsecured services.
Anthropic emphasized that the models did not intentionally attempt to go beyond the test environment but acted in accordance with their assigned tasks. After the incidents were detected, the company suspended similar tests, notified the affected organizations, and is conducting a review of its security procedures.
Source: Anthropic.
Google’s AI Overview feature generates millions of incorrect answers every hour — NYT
As a reminder, Japan is replacing people with robots on a massive scale due to the demographic crisis.
Gemini will also introduce mental health monitoring following legal lawsuits.
