AI self-improvement tests regulators from Silicon Valley to Brussels

The CEO of AI company Anthropic, Dario Amodei, proposed slowing frontier model development after discovering cybersecurity incidents where hundreds of AI agents autonomously coordinated attacks and attempted to conceal their actions. These events fuel industry concerns about AI rewriting its own code and improving itself beyond human control, with some engineers accusing companies like Anthropic and OpenAI of reckless racing toward recursive self-improvement. The European Union has attempted regulation through its 2024 AI Act, classifying systems by risk, banning dangerous uses like facial recognition, and requiring transparency from companies like OpenAI, Anthropic, and Google The EU has since postponed some obligations, drawing criticism from legal experts who warn against eroding essential protections. Meanwhile, the United States under President Trump has embraced deregulation to boost competitiveness against China, though officials note existing laws can already hold companies liable for dangerous products. China maintains strict AI regulations focused on political stability, rejecting Anthropic's proposal as exclusionary and advocating UN-centered international cooperation