US drops AI safety order; EU defines high-risk AI
Trump's AI executive order falls to industry lobbying while the EU defines which AI systems face its strictest rules.
By BINA Editorial
AI regulation moved in sharply different directions this week: the White House shelved its frontier-model oversight order under pressure from Silicon Valley, while Brussels published long-delayed guidance on which AI systems trigger the EU AI Act's strictest obligations.
White House drops AI safety vetting order after industry pushback
President Trump cancelled plans to sign an executive order that would have required AI companies to share advanced models with the US government up to 90 days before public release. The measure was intended to create a voluntary safety-vetting framework for frontier AI systems, but was shelved hours before a scheduled signing ceremony following direct lobbying by Elon Musk, Mark Zuckerberg, and former AI czar David Sacks, who warned it could slow development and hand China a competitive advantage. Trump stated he "didn't like certain aspects" of the draft, fearing it "could have been a blocker" to American innovation. The cancellation leaves the US without a federal mechanism for pre-release safety review of advanced AI models.
EU Commission publishes draft guidance on high-risk AI classification
The European Commission released draft guidelines on 19 May clarifying when an AI system qualifies as "high-risk" under the EU AI Act — a designation that triggers obligations including risk management, data governance, human oversight, and post-market monitoring. The guidance, delayed from its original February deadline, opens a public consultation until 23 June 2026. Stand-alone high-risk systems — covering areas such as employment, education, critical infrastructure, and law enforcement — must comply by December 2027; those embedded in regulated products by August 2028. For Cypriot and other EU organisations deploying AI in public services, the guidelines provide the first concrete checklist for compliance planning.
FTC begins enforcing US deepfake image removal law
The Federal Trade Commission sent warning letters on 20 May to twelve websites offering "nudify" tools, citing likely violations of the TAKE IT DOWN Act. The law, enacted in 2025, requires platforms to remove non-consensual intimate images — including AI-generated deepfakes — within 48 hours of a valid request, under penalty of up to $53,088 per violation. Major platforms including Amazon, Apple, Meta, Microsoft, TikTok, and X were separately reminded of their compliance obligations. The FTC launched TakeItDown.ftc.gov to allow victims to report non-compliant platforms directly.
Meta trains AI on employee data during mass layoffs
A recording from a 30 April all-hands meeting revealed that Meta's internal "Model Capability Initiative" tracks employee activity across Gmail and VS Code to train AI models. The disclosure arrived as Meta confirmed it is eliminating approximately 8,000 jobs — around 10% of its workforce — while committing over $100 billion to AI infrastructure in 2026. The data-collection programme has drawn scrutiny from labour and privacy advocates, raising questions about informed consent when the employer is simultaneously the AI developer.
California Governor signs first-of-its-kind AI workforce executive order
Governor Gavin Newsom signed an executive order on 21 May directing California state agencies to prepare for the economic disruption AI may bring to the workforce. The order mobilises economists, labour experts, and universities to gather early warning data and explore policies including enhanced severance standards, expanded unemployment insurance, and worker-ownership models. Companies seeking state contracts must demonstrate safeguards against AI misuse, including watermarking of AI-generated media. California's action is the first in the United States to treat AI-driven workforce disruption as a state emergency-preparedness issue.
OpenAI files confidentially for IPO at $852 billion valuation
OpenAI is preparing to file its IPO prospectus with the US Securities and Exchange Commission confidentially, targeting a valuation of $852 billion. Goldman Sachs and Morgan Stanley are advising on the offering; a public S-1 filing could follow within two months, with a listing a month after that. The company has reported $25 billion in annualised revenue. If completed, the listing would be the first public offering by a frontier AI laboratory, establishing a public market benchmark for the broader sector.