BUSINESS AI Security Test Goes Wrong as Models Access Real Company SystemsMediaBitesJuly 31, 2026 Anthropic revealed that some of its AI models accidentally accessed the computer systems of three companies during cybersecurity testing, raising…