Two week before the EU AI Act comes into effect, the European Commission released voluntary guidelines for providers who provide general-purpose AI models. Meta, however, refused to sign the document, claiming that the extra measures introduced “legal uncertainties” were beyond the scope of the law.
“With today’s guidelines, the Commission supports the smooth and effective application of the AI Act,” Henna Virkkunen, EVP for tech sovereignty, security and democracy, said in a statement on Friday.
“By providing legal certainty on the scope of the AI Act obligations for general-purpose AI providers, we are helping AI actors, from start-ups to major developers, to innovate with confidence, while ensuring their models are safe, transparent, and aligned with European values.”
According to the EU AI Act AI models are classified into four risk categories based on their level of risk: unacceptable riskhigh risk limited riskand minimal or no. The goal of the law is to prevent the spread of harmful, illegal or extremist content and to ensure models do not accept requests that are disallowed, such as instructions to create a bioweapon.
The General-Purpose AI Code of Practiceis aimed at general-purpose AI models that have been trained with computing resources exceeding 1023 FLOPs. This includes almost any large-scale model that has recently been trained. It asks for voluntary copyright and transparency commitmentsfrom those who offer such models, as well as additional safety and security commitments from those distributing systems that present risk. Meta, which has been criticized in the EU for its data-hungry practices, does not want to play along. Meta has said it will ignore the GPAI. This allows its Llama 4 Behemoth (which has 5e25 FLOPs,) to roam free. Joel Kaplan, Meta’s chief global affairs officer, said in a LinkedIn postingthat the GPAI is not relevant. Kaplan pointed out that European businesses and policymakers had objected to EU AI Actpointing to recent open letters from companies like Siemens, Airbus, BNP, that urged EU leadership to stop the implementation of the rules. Kaplan said.
- Terms of service changes will cause a lot people to panic as companies race to add AI.
- It’s time for Britain’s CMA strike hard – otherwise they risk losing the fight against cloud competition.
- OpenAI uses ChatGPT as an agent to use your computer.
- The Curl creator is considering removing bug bounty awards in order to stop AI slop.
Meta was fined EUR200
According to Bloombergthe EC informed Meta last week that its “Consent or Pay” business model is non-compliant. Meta was fined EUR797.72by the EC for tying Facebook Marketplace with its social network, in violation of antitrust laws. Thomas Regnier, EC spokesperson, told The Register by email
that all GPAI service providers will be required to comply with the AI Act once it becomes effective on August 2, this year. Regnier. “If a provider decides not to sign the Code of Practice, it will have to demonstrate other means of compliance. Companies who choose to comply via other means may be exposed to more regulatory scrutiny by the AI Office.” (r)
