The Mythos moment

Source: economist.com

TL;DR

The story at a glance

The weekly edition spotlights Anthropic's Claude Mythos AI model, which demonstrated unprecedented ability to uncover critical software flaws in banking and infrastructure systems, prompting its restricted access. Leaders like Dario Amodei of Anthropic lead a handful of firms controlling such tech, alarming the Trump administration and sparking calls for oversight. This "Mythos moment" coincides with recent Hungary elections and Iran war tensions, reported now as AI risks shift U.S. policy from hands-off to interventionist. A prior AI safety scare with GPT-2 in 2019 provides context for today's guarded releases.[[3]](https://www.economist.com/science-and-technology/2026/04/15/how-ai-hackers-will-shake-up-cyber-security)[[4]](https://www.economist.com/business/2026/04/08/how-dangerous-is-mythos-anthropics-new-ai-model)

Key points

Details and context

Anthropic, focused on AI safety, trained Mythos as part of frontier models racing ahead; it outperforms human hackers, exposing flaws in widely used operating systems and financial software. The firm argues general release could enable attacks on critical infrastructure, echoing 2019 OpenAI's GPT-2 holdback over misinformation fears—though capabilities inevitably spread via leaks or rivals.[[4]](https://www.economist.com/business/2026/04/08/how-dangerous-is-mythos-anthropics-new-ai-model)

This triggers U.S. rethink: prior laissez-faire aided innovation but ignored dual-use risks like cyber weapons. Trump administration weighs export controls and compute limits, but heavy regulation could entrench big firms' moats, harming competition—especially versus state-backed Chinese AI.[[2]](https://www.economist.com/leaders/2026/04/16/america-wakes-up-to-ais-dangerous-power)

Elsewhere, Hormuz blockade—Trump's bid to squeeze Iran's economy after failed bombings—threatens oil flows (15% global supply) and fertiliser, portending harvests down 10-20% and hunger for millions. Hungary's shift from Orban tests democratic reversal via anti-corruption.[[5]](https://www.economist.com/leaders/2026/04/15/how-to-end-the-war-in-iran)

Key quotes

"Anthropic’s latest creation is so startlingly good at finding software vulnerabilities that, in the wrong hands, it would threaten critical infrastructure." — The Economist leader.[[2]](https://www.economist.com/leaders/2026/04/16/america-wakes-up-to-ais-dangerous-power)

"Should a handful of men be entrusted with the world’s most potent new technology?" — Print edition cover framing the Mythos moment.[[2]](https://www.economist.com/leaders/2026/04/16/america-wakes-up-to-ais-dangerous-power)

Why it matters

AI's hacking prowess exemplified by Mythos elevates it from economic tool to strategic weapon, demanding global safeguards akin to nuclear non-proliferation. For businesses and governments, restricted access means elite firms gain defensive edges but risks a bifurcated market where only giants thrive, while consumers face delayed benefits and potential job displacement. Watch U.S. regulatory moves like model approvals or chip curbs, and rivals' responses—though open-source leaks could undermine controls soon.[[6]](https://www.economist.com/by-invitation/2026/04/19/no-to-laissez-faire-on-ai-yes-to-a-light-touch)

[[1]](https://www.economist.com/weeklyedition/2026-04-18)[[2]](https://www.economist.com/leaders/2026/04/16/america-wakes-up-to-ais-dangerous-power)