The Biggest AI Announcements of May 2026

May 2026 was one of the denser months for AI news in recent memory. Three major model releases, a significant acquisition, and the first formal enforcement actions under the EU AI Act all landed within the same four-week window.

Model releases

Google released Gemini 2.5 Ultra with a claimed 1 million token context window and improvements on multimodal reasoning tasks. Early benchmarks are competitive with GPT-5 on coding and maths, though human preference evaluations favour GPT-5 and Claude 4 for natural language tasks. Meta followed with Llama 4.1, an open-weights model that narrows the gap with frontier models significantly — particularly for code and structured outputs.

Business and regulation

The EU's AI Act enforcement phase began in earnest, with the first formal assessments of high-risk AI systems required from companies operating in Europe. The Act's requirements around transparency, human oversight, and bias testing are pushing enterprise AI vendors to document their models more carefully.

On the acquisition front, Adobe completed its purchase of a leading AI video startup, signalling continued consolidation in the creative AI space. The deal values AI-native creative tools at multiples that would have seemed unrealistic two years ago.

Research worth reading

• 'Scaling Laws Are Not Enough' — DeepMind, arguing that raw model scale yields diminishing returns without architectural innovation
• 'Constitutional AI at Scale' — Anthropic's update on how RLHF training methods are evolving for frontier models
• Stanford HAI's report on AI in healthcare, showing measurable diagnostic accuracy improvements with appropriate human oversight