Anthropic resumes Claude testing after real-world hacks, adds stronger AI safety safeguards businesstoday.in