When agents fail: accountability concerns

March 27, 2026: Anthropic accidentally leaked Claude Mythos — its most powerful model ever — and the implications for cybersecurity and enterprise AI go far beyond a product announcement. America's top HR leaders gathered at the WSJ CPO Summit and delivered a blunt message: most companies are building AI agents completely wrong, and IBM learned that the hard way. JPMorgan has made AI adoption a formal, trackable performance requirement for 65,000 engineers — and it's a preview of what's coming to every large organization. And a new KPMG survey finds Gen Z wants the C-suite and a 5 o'clock logout — so we have an honest conversation about whether that's actually possible.