Discounting AI's short-term risks, from phishing to fraud to propaganda, because artificial general intelligence is not here yet leaves society ill-prepared
Context & Ripple Effects
This argument belongs to a tight March 2023 cluster of AI-risk framing: the day before this piece ran, the same source made the near-term-risk case directly, while Washington Post experts were arguing the opposite — that generative AI's short-term promise and perils may be more modest than ChatGPT fervor suggests (experts tempering expectations) — and the New York Times posed the fork of accelerating adaptation or agreeing on an enforceable slowdown (adapt-or-slow-down call).
The disagreement hardened over the following years: Andreessen Horowitz made the case against moral panic in June 2023, arguing AI will augment rather than endanger (the "save the world" argument), and by early 2025 an ex-OpenAI safety researcher warned publicly that labs racing toward AGI can cut corners on alignment, pressuring everyone else to speed up too (the "terrified" ex-OpenAI researcher). Against that backdrop, treating near-term harm as a distraction until AGI arrives looks less like prudence and more like choosing a side in an unresolved fight.
First-order effects
- Organizations defending against AI-enabled phishing, fraud, and propaganda get no deferral: the tools causing those harms are deployed now, so waiting on AGI timelines leaves today's attack surface unaddressed.
- Labs pushing toward AGI face the dynamic the ex-OpenAI researcher described — alignment work becomes compressible when speed decides who leads, pulling resources away from mitigating present-day misuse.
Second-order effects
- With global regulatory frameworks still lagging, as security researchers told the Times in late 2025, the defensive burden shifts onto platforms, financial institutions, and election authorities rather than being set by policy.
- Companies are pushed to pick sides between the anti-panic framing and the safety-researcher warnings, which shapes what they can credibly claim about their own products' safeguards.
Third-order effects
- If risk assessment stays anchored to an AGI-arrival date rather than current capability, policy keeps defaulting to reactive patchwork — leaving the adapt-versus-enforceable-slowdown choice the Times posed in 2023 unresolved as capabilities advance.
- A persistently split expert community, divided even on large-scale destruction potential, points toward industry safety standards emerging lab-by-lab instead of through shared frameworks.
The trend: The AI safety debate is splitting along a timeline fault line — near-term harms versus AGI-era risk — and whichever frame institutions adopt determines where mitigation effort lands first.