OpenAI and Anthropic unite against open-weight AI risks to their bottom line

Original Article ↗ Hacker News Discussion ↗

US AI giants OpenAI and Anthropic are urging regulators to crack down on open-weight models, especially those coming from China, arguing they pose safety and security risks once weights are irreversibly released. Commenters largely see this as an attempt at regulatory capture to protect fragile, capital-intensive business models built on copyrighted training data, warning it could damage US competitiveness and accelerate a shift toward Chinese or other foreign open models. Underneath is a broader fight over whether AI should be treated as a tightly controlled, proprietary technology or a widely accessible, open infrastructure, and what that choice means for innovation, security, and economic power.

Regulation, Monopoly, and Motives

  • Many see OpenAI/Anthropic’s stance as classic regulatory capture: using “safety” to protect valuations, moats, and overextended investors.
  • Comparisons to US antitrust history and “consumer welfare” rhetoric: regulation seen as designed to protect incumbents rather than citizens.
  • Some argue this will regulate the US into uncompetitiveness, echoing tariffs on Chinese EVs and past protectionism.

US vs China and Geopolitics

  • Strong theme: US trying to constrain open models while China aggressively releases capable open-weight models.
  • Some expect the US to pressure EU and others to follow, but others note partners are already diversifying away from US tech and finance.
  • A minority argues US firms, though profit‑maximizing, are more trustworthy than Chinese entities with explicit geopolitical motives; others dispute this.

Open-Weight vs Closed Models

  • Many commenters are strongly pro open-weight, seeing it as essential for:
    • Competition and avoiding winner‑take‑all outcomes.
    • Security work (Hugging Face incident cited: closed models’ guardrails impeded response).
    • Global access outside the US.
  • Critics warn that uncensored frontier open models could enable cyberattacks, bio/weapon guidance, and other mass harm.

Distillation, IP, and Copyright

  • Heated debate on “distilling” from proprietary APIs:
    • One side: it’s fair use / querying an API; closed labs themselves trained on massive scraped and copyrighted data.
    • Other side: this is straightforward IP theft, especially when targeting specific “thinking tokens.”
  • Settlements over copyright scraping are cited as evidence labs already “stole” data yet keep the gains.
  • Underlying disagreement about whether copyright is a moral right, a pragmatic incentive system, or already broken by overlong terms.

Economic & Industry Impacts

  • Disagreement on economic effects:
    • Some say distills compress margins and slow frontier R&D, hurting US leadership.
    • Others argue cheaper open models boost global productivity and the broader US economy, even if they hurt a few big labs.
  • Several note AI/data-center spending is propping up US growth; that gives government a strong incentive to side with incumbents.

AI Safety, Risk, and Community Dynamics

  • Deep split:
    • One camp dismisses current “AGI risk” talk as overblown marketing and fearmongering; points to repeated “too dangerous to release” claims that didn’t pan out.
    • Another insists those working on frontier models see real systemic risks, especially if open access coincides with rapid capability jumps.
  • Meta‑discussion that HN becomes polarized and toxic whenever AI safety or regulation is discussed.