Report from TechCrunch
In Brief – The European Commission has announced the release of the final draft AI Code of Practice that gives “voluntary” guidelines to AI developers to help them comply with AI regulation under the EU’s AI Act. The code focuses on copyright protection, transparency into the development and operation of AI systems, and “safety”. AI developers are not required to abide by the guidelines, but the Commission, which will be phasing in enforcement of the AI Act over the coming two years, is encouraging adoption by saying that AI developers who agree to the practices could benefit from a “reduced administrative burden and increased legal certainty,” while those that do not may find proving compliance more costly. Among the most controversial aspects of the code are its copyright provisions, which include asking AI developers to never pirate materials for training, allow rights holders to opt their content out, abide by technical efforts to block content gathering web-crawlers, and share detailed information about the composition and collection of their training data. The code’s safety guidance calls for company monitoring of numerous harms, including cybersecurity breaches, disruptions of critical infrastructure, and serious harm to mental or physical health, with timelines to report serious incidents to the EU’s AI Office. The code must be approved by the European Commission and EU member states amid ongoing industry pushback.
Context – Following the release of Chat-GPT, the European Parliament changed course on the AI Act and pushed to regulate core “foundation models”. This shift was divisive among EU-based AI innovators, and concerns have only grown. The US, UK, and Japan have since focused on encouraging AI investment rather than regulation. Some European government leaders, backed by the Draghi Report critical of over-regulation in Europe, have joined in, and top EU business and tech leaders are calling for an AI Act delay. But EU parliamentarians and civil society groups have pushed back, and the emerging discrepancy over AI training and copyright law between the US and EU may prove particularly significant.