Breaking news and updates from the world of technology.
MAI-Thinking-1, announced at Microsoft Build on June 2, is a 35-billion-parameter reasoning model that independent evaluators preferred over Claude Sonnet 4.6 for overall quality. It hits 52.8% on SWE-Bench Pro. It was trained without OpenAI data.
A new White House executive order asks AI developers to voluntarily submit their most powerful models for government cybersecurity review up to 30 days before public release — a scaled-back version of an earlier proposal that industry groups had pushed back against.
The UK's Competition and Markets Authority has issued binding requirements on Google, ordering the company to let publishers opt out of AI Overviews and model fine-tuning — the first time any major regulator has imposed these controls on AI search.
OFAC designated Nobitex, Wallex, Bitpin, and Ramzinex — exchanges that together processed the majority of Iran's digital asset inflows in 2025 — for IRGC financing, ransomware facilitation, and sanctions evasion.
A new White House directive requires AI developers to submit their most powerful models for government review before release, triggered by alarm over AI's vulnerability-discovery capabilities.
Florida Attorney General James Uthmeier filed an 83-page civil lawsuit against OpenAI and CEO Sam Altman on June 1, accusing the company of concealing safety risks, collecting minors data without parental consent, and ignoring expert warnings while marketing ChatGPT to children. Florida is the first state to take such legal action.
Google's parent company announced its largest ever equity raise, setting 2026 capital expenditure guidance at $180-$190 billion as AI demand outpaces available compute supply.
The AI safety company closed a $65 billion Series H round last week and is now targeting a public listing in fall 2026, at a valuation that would rank it among the most valuable companies ever to go public.
Verizon's 2026 Data Breach Investigations Report marks a historic shift: vulnerability exploitation now accounts for 31% of breaches, overtaking stolen credentials for the first time in 19 years. Median patch time increased to 43 days. Shadow AI tripled. AI is compressing exploitation windows to hours.
A new Aithos study using the LARA framework tested 12 frontier AI models against GDPR and EU AI Act requirements. The best model still failed 46% of scenarios. The worst failed 93%. And it is the businesses deploying these models, not the AI companies, who face the fines.
The Commerce Department's May 31 guidance closes the overseas subsidiary workaround that allowed Chinese companies to procure NVIDIA and AMD's most advanced AI chips. NVIDIA derived over 20% of its compute revenue from China through intermediary channels -- that channel is now closed.
Announced at the Choose France summit, SoftBank's investment covers 5 gigawatts of AI compute capacity across three sites in Hauts-de-France. The EUR 45 billion firm tranche targets completion by 2031, with EDF, Schneider Electric, and SB Energy as partners.