AI Regulation and the Regulated Enterprise — Trajectory to 2030

Narrative

The period from early 2023 to mid-2026 saw rapid advancements and releases from frontier AI labs, alongside increasing efforts in external evaluation and safety. OpenAI continued to iterate on its GPT series, releasing GPT-4 in 2023 with enhanced capabilities and later introducing GPT-4o in August 2024, followed by a series of GPT-5.x models in 2025 and 2026, including GPT-5.4 as a capable frontier model for professional work and GPT-5.5 Instant for improved conversational AI. Anthropic also progressed its Claude family, launching Claude 3 models (Haiku, Sonnet, Opus) in March 2024 with multimodal input and expanded context windows, culminating in the release of Claude 4 (Opus 4, Sonnet 4) in May 2025, which brought significant improvements in coding, reasoning, and autonomous task execution.

Google DeepMind focused on its Gemini series, releasing Gemini Nano and Pro in December 2023 and Gemini Ultra 1.0 in February 2024. By late 2025 and early 2026, Gemini 3 emerged with advanced multimodal understanding and reasoning, with Gemini Deep Think demonstrating the ability to solve professional research problems in mathematics and science. Meta AI contributed to the open-source landscape with the LLaMA series, introducing LLaMA 3 in July 2024, which offered multilinguality, coding, reasoning, and tool usage, and later LLaMA 4 in April 2025, featuring a sparse Mixture-of-Experts architecture and extended context windows.

Mistral AI rapidly expanded its offerings, with Mistral Large 24.11 and Codestral 25.01 announced in late 2024 and early 2025, focusing on reasoning, coding, and long context. By December 2025, Mistral 3 was released, including a sparse mixture-of-experts model and smaller dense models, emphasising open-source and efficient AI. xAI, founded by Elon Musk in July 2023, launched its Grok series, with Grok 3 released in February 2025, trained on the Colossus supercomputer, and Grok 4.20 Reasoning and Non-Reasoning variants appearing in March 2026, notable for real-time data access and a distinctive tone.

External evaluations from METR (Model Evaluation & Threat Research) provided crucial independent assessments of these frontier models. METR published reports on GPT-4 and Claude in March 2023, and continued to evaluate subsequent releases such as GPT-4o, Claude 3.5 Sonnet, GPT-4.5, Claude 3.7, GPT-5, and GPT-5.1-Codex-Max through 2024 and 2025. In January 2026, METR released Time Horizon 1.1, an updated methodology for measuring AI autonomous capabilities, which indicated an accelerated rate of progress in AI capabilities since 2023. The Frontier Model Forum, established in July 2023 by major labs, also published research on AI safety frameworks and risk management.