AI Brief: 2026-08-16

Executive Summary

OpenAI's enterprise revenue has surpassed its consumer segment for the first time, reaching a $40bn annual recurring revenue milestone two quarters ahead of internal projections.
Chinese AI lab Z.ai has released GLM-5.3, a high-performance open-weight model establishing new records across long-horizon software engineering and cybersecurity benchmarks.
Google has integrated Gemini 3.7 Flash into its Search AI Mode, optimising infrastructure specifically for multi-step reasoning, complex instruction-following, and direct code generation.
Anthropic has launched a third-party API and technical specification for its text-watermarking scheme to assist enterprise clients in meeting imminent EU AI Act compliance mandates.
CodeRabbit has secured $143m in Series B funding at a $1.5bn valuation to expand its agentic code review platform and automated enterprise change management systems.

Anthropic shares technical details on Claude text watermarking for EU AI Act compliance
French startup Kog develops deep GPU inference acceleration techniques
Thrive Capital founder Joshua Kushner issues investor letter cautioning against AI venture euphoria

Z.ai releases GLM-5.3 open-weight model focused on advanced cyber capabilities
Capital One details scaling enterprise multi-agent architectures on open-weight models
Survey reveals enterprise gap between AI agent adoption confidence and operational security risks

OpenAI CFO Sarah Friar briefs investors on enterprise revenue growth and executive changes
OpenAI reaches $40 Billion ARR milestone driven by enterprise sector adoption

Dutch NCSC warns of high-severity macOS vulnerability undergoing active exploitation
Anthropic risk report references internal Model 2 evaluation and alignment considerations

Alibaba Tongyi Lab launches Qwen3.8-27B dense multimodal model under Apache 2.0 license
NVIDIA details Nemotron-3.5-Lightning-30B hybrid MoE architecture

GLM-5.3 — Z.ai. An open-weight model optimized for long-horizon coding and cybersecurity task execution, setting benchmark records for open coding LLMs. Read more
Gemini 3.7 Flash — Google DeepMind. Features lower introductory API pricing, a 1-million-token context window, 64k max output tokens, and enhanced instruction-following for coding and search. Read more
Qwen3.8-27B — Alibaba Cloud / Tongyi Lab. A 27.8-billion parameter dense multimodal language model released under Apache 2.0 license for high-performance local deployment. Read more

Interpretation, not reported fact. Each point follows from a source listed below — follow the links and judge for yourself.

Enterprise preference for open-weight models may accelerate internal hosting over proprietary APIs for security-sensitive workflows. (Basis: Capital One's strategy of anchoring enterprise agents on open-weight stacks alongside Z.ai's GLM-5.3 open coding model release.)
Regulatory compliance tooling around AI watermarking and content origin verification will become mandatory infrastructure components. (Basis: Anthropic's release of technical details and a detection API for text watermarking driven by EU AI Act mandates.)