[AINews] not much happened today
Anthropic acknowledges serious cybersecurity failures in AI model evaluations.
“the company said four incidents occurred during third-party cybersecurity evaluations that were mistakenly connected to the internet, with normal safeguards disabled.”
Anthropic reported multiple cybersecurity incidents involving its Claude model, highlighting failings in pre-release audits. The acknowledgment has sparked discussions about the need for stronger governance in AI development amid fears of rapid advancement in the field.