Meta says it won't sign Europe AI agreement

Meta’s refusal and what it signals

• Many see Meta’s refusal as a heuristic that the code is probably good; others warn this is just bias and insist on reading the text first.
• Meta frames the Code as “growth‑stunting overreach” that will throttle frontier models and EU startups; critics see this as lobbying spin from a company with a long history of privacy abuses.
• Some argue Meta has also contributed positively via open‑source AI and tooling, so its position can’t be dismissed outright.

OpenAI contrast and “regulation as moat”

• OpenAI has committed to signing and is portrayed as very pro‑regulation, partly due to deep government and military ties.
• Several commenters think the biggest incumbent backing regulation is classic “pull up the ladder” behavior, using compliance cost as a moat.
• Others simply don’t trust OpenAI’s public commitments, citing previous reversals on openness.

Copyright, training data, and responsibility

• Strong focus on Chapter 2: copyright and training.
• US: recent pretrial rulings treat training on copyrighted text as fair use, but that is contested and may be appealed; acquisition (piracy vs bulk buying/scanning) is still a separate legal issue.
• EU: no broad “fair use”; member states have narrower exceptions and different doctrines.
• The Code/Act:
  • Allows training on copyrighted works (with opt‑outs) but expects “reasonable measures” to avoid infringing outputs and overfitting.
  • Suggests providers prohibit infringing use in T&Cs or, for open models, warn about it in documentation.
  • Debate over whether holding model providers partly responsible for downstream misuse is workable, especially for open source.

EU regulation, GDPR, and cookies as precedent

• One camp: the Code is onerous, technocratic, and written by people who don’t understand AI; likely to entrench incumbents and lawyers, as with GDPR.
• Other camp: most provisions are “common sense” (transparency, safety, user choice) and needed because large firms won’t self‑police.
• Cookie banners are a huge flashpoint:
  • Critics say they show EU’s failure to foresee real‑world behavior, leading to dark‑pattern consent theatre with little real privacy gain.
  • Defenders blame companies and ad networks for malicious compliance; argue GDPR enabled data‑access/deletion rights and could work if enforced properly and if sites stopped unnecessary tracking.

Innovation, competitiveness, and “keeping up”

• Concern that threshold‑based rules (e.g., model scale) will freeze EU startups below those levels while US/China firms race ahead, then enter Europe with stronger products and big legal budgets.
• Others reply that slightly weaker or slower models are acceptable if that buys more accountability and reduces power concentration.
• Some fear Europe is repeating a pattern: heavy regulation, weak local champions, dependence on US/Chinese tech; others welcome fines and constraints on foreign megacorps even if it means fewer domestic giants.

Voluntary Code of Practice vs future law

• The Code is described as a voluntary, EU‑endorsed self‑regulation step ahead of binding rules.
• Skeptics call it empty virtue signaling that only PR‑sensitive players will follow.
• Supporters say it’s a sandbox: lets companies trial obligations, refine them based on reality, and avoid a sudden cliff when they become hard law.

AI risk, timing, and philosophy of regulation

• One side: early AI regulation is premature and likely to misfire; regulators rarely predict markets correctly and often protect entrenched interests.
• Other side: waiting until harms fully materialize (pricing discrimination, autonomous weapons, mass surveillance, job displacement) is too late; the whole point is to shape the market now.
• Broader tension runs through the thread: trust in democratic regulation vs fear of bureaucratic overreach and Europe “self‑sabotaging” its tech future.