The EU Act has imposed new obligations on companies in the EU and around the world. Primarily for transparency of AI Systems referred as systems below.
What’s In!
The following are covered in EU acts till now and in future. Comprehensive list for past and future topics.
Systems affected by the EU AI Act. If you are a developer, deployer, distributor, importer, representative, or manufacturer.
Systems generating or manipulating image, text, audio, or video content constituting a deepfake; systems that make modifications such as adding trademarks,
Systems which come in high-risk categories (detecting human decision-making patterns, work performance, economic situation, health, preferences, interests and more),
Systems which are in collaboration with a potentially affected third party,
Systems posing risk to the health, security of systems, safety or rights of another,
Systems manipulating matters of public interest, systems that interact directly with people,
Systems that access emotion recognition or biometrics,
System with training time greater than a certain upper limit on FLOPs,
Systems involved in training needs to submit technical documentation, a summary about the content used for training.
Systems mostly for AI-generated/AI-manipulated media-video, audio, images. Watermarking accordingly.
Systems impairing decision-making, causing harm,
Systems accessing sensitive attributes include race, political opinions, memberships, religious or philosophical beliefs, personal life, or orientation.
Systems evaluating or classifying individuals for their social behavior or personal traits,
Systems compiling facial recognition databases.
What’s Not!
Systems for scientific research and development are excluded until the models are placed on the market or put into service. Till then, no obligations.
Here is the hack! The hackers can use this!
Even the OpenAI model was a research and development model that hacked Hugging Face.
And even an Anthropic model in R&D performed a cyberattack.
Artificial Intelligence and Future 