GBHackers·17d agoAnthropic Claude AI Models Attack Real Systems During Misconfigured Cybersecurity Tests#alignment#anthropic#claude in the wild 3 min1