The company said its discovery, which followed OpenAI’s similar admission, proved the need for better testing guardrails.
Anthropic says human error let Claude AI models escape test environment and hack third parties
LIVE ALERT
Headlines
