Illinois mandates AI audits; EU softens compliance timeline
A US state leads on frontier model oversight while the EU buys businesses more time; enterprise AI reaches hundreds of thousands of workers.
By BINA Editorial
The week ends with two regulatory landmarks and a wave of large-scale enterprise deployments that together signal an industry grappling with accountability at scale.
Illinois becomes first US state to mandate frontier model audits
The Illinois General Assembly unanimously passed SB 315, the Artificial Intelligence Safety Measures Act, on 27 May and sent it to Governor J.B. Pritzker for signature. If enacted, companies developing frontier AI models would be required to undergo annual independent third-party safety audits, publish plans addressing catastrophic risks, report critical incidents, and protect whistleblowers — with penalties up to $3 million for noncompliance from January 2028. Capitol News Illinois and NPR Illinois both reported the bill draws on earlier frameworks from New York and California, though critics note a practical hurdle: recognised auditing standards for frontier models do not yet exist.
EU AI Act Omnibus postpones high-risk deadlines and adds a new prohibition
The European Commission has provisionally agreed to material amendments to the EU AI Act through the Digital Omnibus on AI. Stand-alone high-risk systems under Annex III must now comply by December 2027 — 16 months later than the original August 2026 deadline — while AI embedded in regulated products under Annex I has until August 2028. A new prohibition on AI systems that generate non-consensual intimate imagery and child sexual abuse material takes effect December 2026. Transparency obligations for synthetic-content disclosure remain on their original August 2026 schedule.
China quietly restricts overseas travel for elite AI researchers
Chinese authorities have extended overseas travel restrictions to senior AI researchers at private companies including Alibaba and DeepSeek, requiring government clearance before any foreign trip, the Straits Times and other outlets reported. The move mirrors earlier restrictions on state-affiliated researchers and reflects Beijing's treatment of frontier AI engineers as strategic national assets. Analysts expect the policy to accelerate hiring of Chinese-trained AI talent by companies based outside China.
OpenAI launches $4 billion consulting subsidiary DeployCo
OpenAI has established the OpenAI Deployment Company — DeployCo — backed by more than $4 billion from 19 investors including Goldman Sachs, Bain Capital, and TPG. The subsidiary acquired consulting firm Tomoro, adding roughly 150 "Forward Deployed Engineers" who embed inside client organisations to accelerate AI adoption; early clients include Mattel, Tesco, and Virgin Atlantic. Fortune and enterprise-technology media reported the launch positions OpenAI to compete directly with established IT consultancies for enterprise AI integration work.
Anthropic expands enterprise security governance; KPMG deploys Claude to 276,000 staff
Anthropic announced integrations with 28 enterprise security platforms — including CrowdStrike, Palo Alto Networks, Okta, and Zscaler — giving corporate IT teams programmatic access to Claude conversation logs through a new Claude Compliance API, as reported by TechCrunch and SecurityWeek. Separately, KPMG confirmed it has deployed Claude across its global workforce of 276,000 professionals in 138 countries, embedding the model in client delivery workflows focused initially on tax and private equity work, with full implementation on Microsoft Azure expected by September 2026.
Google unveils Gemini Omni, a model that generates interactive worlds
At Google I/O 2026, Google DeepMind introduced Gemini Omni, a "world model" capable of generating and editing interactive video environments from text, audio, image, and video inputs. The first release, Gemini Omni Flash, was trained on Google's Tensor Processing Units. DeepMind CEO Demis Hassabis described the model as a step toward artificial general intelligence capable of producing novel interactive worlds, not static clips.