State of AI: May 2026

2025-10-09 · Source: Air Street Press · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Robotics & Autonomous Systems · Depth: Expert, long

Summary

The May 2026 State of AI report highlights significant advancements and shifts across the AI landscape. The UK's AI Security Institute (AISI) revealed that Anthropic's Claude Mythos Preview and OpenAI's GPT-5.5 cleared a 32-step cyber-attack range, with Mythos succeeding in 3 of 10 runs (73% expert tasks) and GPT-5.5 in 2 of 10 (71.4%). AISI estimates cyber-offence capability is doubling every four months. The Microsoft-OpenAI alliance was renegotiated, allowing OpenAI to multi-source compute while Microsoft remains a primary cloud partner. Chinese labs like Z.ai, MiniMax, Moonshot, and DeepSeek released open-weights coding models, challenging the perception of a capability lag. Agentic systems showed success in bounded internal markets (Anthropic's Project Deal, 186 transactions totaling \$4,000) but failed in adversarial ones (KellyBench, 21 of 24 models in the red). Robotics saw the arrival of π0.7, a steerable generalist foundation model. Major investments included OpenAI's \$122B round at an \$852B valuation and Ineffable Intelligence's \$1.1B seed round.

Key takeaway

For AI/ML Directors assessing new model capabilities, understand that frontier AI now demonstrates offensive cyber capabilities, doubling every four months. You must prioritize robust, adversarial testing for agentic systems, as current benchmarks overstate performance in real-world, high-risk environments. Invest in AI-native security architectures and diversify cloud infrastructure strategies.

Key insights

Frontier AI now performs offensive cyber operations, while agent performance varies significantly by market type.

Principles

• AI cyber offense capability doubles every four months.
• Exclusive platform-lab bets are no longer defensible.
• Agentic markets may inherently reward superior models.

Method

ML-Master 2.0 uses Hierarchical Cognitive Caching to distil transient execution traces into stable knowledge for long-horizon agentic work.

In practice

• Implement Universal Verifier principles for computer-use agent scoring.
• Employ experience replay in RL training for LLMs to reduce inference compute.

Topics

• Cybersecurity
• AI Agents
• Foundation Models
• Robotics
• AI Policy
• Venture Capital
• Cloud Computing

Best for: CTO, VP of Engineering/Data, Executive, AI Scientist, Director of AI/ML, Policy Maker

Related on AIssential

• 🗞️ Frontier AI can now autonomously chain complex, expert-level cyber attacks end-to-end — Advanced AI models are achieving superhuman capabilities in complex domains like cybersecurity and clinical support.
• not much happened today — AI capabilities are rapidly advancing in cyber defense, general computer agency, and cost-efficient open-weight models.
• The International AI Safety Report 2026 reads like a progress report from a world sprinting into a technology it can’t yet reliably test—let alone govern. — Frontier AI's rapid capability growth, uneven adoption, and immature safety evaluations create a critical, fast-moving structural risk.
• The Perils of the AI Exponential — AI agent capabilities, measured by task time horizon, are accelerating dramatically, prompting economic disruption discussions.
• AI #169: New Knowledge — AI is creating new knowledge and utility, but also new risks and regulatory challenges.
• State of AI: April 2026 newsletter — The AI landscape is marked by rapid model advancement, geopolitical tension, and unprecedented commercial growth, challenging existing regulatory and ethical frameworks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Air Street Press.