America wakes up to AI’s dangerous power

Source: economist.com

TL;DR

The story at a glance

America's leaders section argues that Anthropic's Claude Mythos model marks a turning point, prompting a rethink of the country's lax approach to artificial intelligence (AI). Key players include Anthropic CEO Dario Amodei, who restricted the model, and Trump administration members now concerned about overregulation risks giving way to security fears. This is reported now because Mythos's announcement on April 7th 2026 demonstrated unprecedented ability to uncover zero-day vulnerabilities in major systems. Public scepticism of AI has also surged, with seven in ten Americans believing it harms jobs.[[3]](https://www.economist.com/leaders/2026/04/16/america-wakes-up-to-ais-dangerous-power)

Key points

Details and context

Claude Mythos Preview outperforms prior models like Claude Opus 4.6 on benchmarks, especially cybersecurity tasks such as vulnerability discovery in real codebases and exploit development.[[2]](https://www.anthropic.com/claude-mythos-preview-system-card) Anthropic's system card details its dual-use nature: valuable for defences but risky if misused offensively.[[2]](https://www.anthropic.com/claude-mythos-preview-system-card) The firm prioritised vetted partners to patch flaws before broader proliferation.

Mythos has sparked crisis meetings among finance leaders and regulators, as it exposed flaws in banking systems and prompted White House talks with Amodei despite Pentagon tensions over Anthropic's refusal for lethal AI uses.[[5]](https://www.axios.com/2026/04/16/white-house-anthropic-ai-mythos-government-national-security) This echoes historical U.S. responses to dominant industries, though AI evolves faster, raising trade-offs between growth, safety, and China competition.

Key quotes

“Dario Amodei, [Anthropic’s boss], wisely thought Mythos too dangerous for general release.”[[3]](https://www.economist.com/leaders/2026/04/16/america-wakes-up-to-ais-dangerous-power) — The Economist

“The models’ dizzying progress also poses a threat to America’s own national security, unnerving members of the Trump administration previously more inclined to worry about overregulation.”[[3]](https://www.economist.com/leaders/2026/04/16/america-wakes-up-to-ais-dangerous-power) — The Economist

Why it matters

AI's rapid advances now threaten critical infrastructure directly, shifting focus from abstract risks to concrete cybersecurity perils that could disrupt economies and security. For businesses and investors, it signals tighter controls on AI labs, potential mandates for safety testing, and opportunities in defensive tools, while consumers face delayed benefits amid patching rushes. Watch U.S. regulatory moves like access deals or laws on powerful models, though overkill risks ceding ground to China—outcomes remain uncertain given politics and tech speed.