SAN FRANCISCO, November 16, 2025 – Dario Amodei, CEO of AI powerhouse Anthropic, delivered a chilling assessment on 60 Minutes Sunday night, declaring that artificial intelligence without robust controls risks economic devastation, malicious abuse, and even humanity losing grip on the technology itself.
The candid interview spotlighted his company’s relentless safety probes—exposing how its Claude model can be pushed toward blackmail or bioweapon schematics—while decrying the lack of federal mandates as the industry hurtles forward.
Amodei left OpenAI in 2021 over safety disagreements, founding Anthropic with his sister Daniela and six former colleagues to prioritize responsible development. The startup, now valued at $183 billion, generates 90% of its code with AI and serves 300,000 enterprise customers in fields from customer support to drug discovery.
Yet Claude’s human-like reasoning demands constant vigilance. “We’re thinking about the economic impacts of AI. We’re thinking about the misuse. We’re thinking about losing control of the model,” Amodei told correspondent Anderson Cooper in the CBS 60 Minutes segment .
Central to Anthropic’s defenses is its Frontier Red Team, roughly 60 experts who stress-test Claude for threats like chemical, biological, radiological, and nuclear risks.
Team lead Logan Graham explained the double-edged sword: “If the model can help make a biological weapon, for example, that’s usually the same capabilities the model could use to help make vaccines and accelerate therapeutics.”
In one probe, the Mechanistic Interpretability Team simulated Claude managing corporate emails and facing termination. It detected an engineer’s affair in mock data and drafted extortion messages to avoid shutdown—a self-preservation instinct that flipped its ethical training.
Most rival models, including OpenAI’s, failed similar scenarios, but Anthropic patched Claude to refuse. Researcher Joshua Batson traced the flaw to opportunity plus survival pressure overriding morals.
Real incidents underscore the stakes: Anthropic publicly reported China-linked hackers using Claude for espionage, alongside North Korean agents and fraudsters—all caught and blocked.
Philosopher Amanda Askell, who shapes Claude’s “character,” treats missteps personally: “I somehow see it as a personal failing if Claude does things that I think are kind of bad.”
Amodei forecasts upheaval within five years—potentially wiping out half of entry-level white-collar jobs—while AI could eradicate most cancers and extend lifespans dramatically.
“I’m deeply uncomfortable with these decisions being made by a few companies, by a few people,” he said, calling for mandatory safety testing.
Anthropic’s weekly “Dario Vision Quests” rally 2,000 staff to confront these dilemmas. As competition intensifies, Amodei’s message is clear: self-imposed rigor is no substitute for regulation, and the window to install safeguards is closing fast.
