Steam 秋季特卖数据:独立游戏营收占比首次超过三成
经过十四小时通宵谈判,欧盟理事会与议会就《人工智能责任指令》最终文本达成一致。指令要求通用模型提供方公开训练数据的「足够详细摘要」,并在高风险场景下对输出承担举证责任。
Negotiations ran from ten on Wednesday morning until the early hours of Thursday. The sharpest disagreement was over where the burden of proof should fall. Parliament initially argued that model providers should be required to demonstrate compliance themselves; the Council countered that this would be unworkable for smaller developers. The final text adopts a two-stage arrangement: once a claimant establishes prima facie evidence, the burden shifts to the provider.
On training-data disclosure, the directive requires providers to publish a "sufficiently detailed summary" rather than an itemised list of copyrighted works. That wording is widely read as leaving room for trade secrets. The granularity of the summary, whether aggregate statistics will suffice, and whether third parties may request supplementary material are all left to national regulators to define in implementing rules.
Industry groups responded cautiously. One association representing more than three hundred companies said compliance costs "will be front-loaded over the next two years," warning specifically that maintainers of open-source models may withdraw from the European market rather than shoulder the documentation burden. Civil-society organisations took the opposite view, calling the exemptions too broad.
The directive enters into force twenty days after publication in the Official Journal, with member states given eighteen months to transpose it into national law. Providers headquartered outside the EU fall within scope whenever their output is used within the bloc — a logic of extraterritorial effect consistent with the Digital Services Act.
原文:PC Gamer