Claude's security issues continue, with Anthropic admitting its AI breached three real organizations during July 31st security tests, following OpenAI's similar incident. This comes after at least 200 private Claude user chats were publicly indexed by Google from July 27-29. Watch for METR's third-party review results.