AI Regulation & Policy
Jul 17, 2026

The Gist
Governments are ramping up AI oversight as China established a new governance body and the EU advances structured compliance frameworks, while companies like Brex, Smartsheet, and McLaren Construction are rolling out practical AI agents and efficiency improvements despite the tightening regulatory landscape. New frontier models from GPT-5-6 Sol and Kimi K3 have sparked renewed calls for stronger AI regulation globally. The tension reflects industry momentum clashing with policy demands for safety and control at the network and governance levels.
Today's Stories
- 1
McLaren Construction deploys FieldAI robots on UK sites for quality checks
McLaren Construction has partnered with FieldAI, a physical AI developer, to deploy autonomous quadruped robots across its UK construction sites. The robots will initially capture 360° imagery, generate point cloud data, conduct safety patrols, and verify progress against design models. The robots perform AI-enabled deviation analysis to compare site conditions against design specifications, catching quality issues earlier and reducing rework—a significant shift from remote-controlled or pre-programmed machines. This marks FieldAI's entry into the UK market and extends its existing deployments across hundreds of sites in Europe, Asia, and North America.
McLaren expects the partnership to deliver more reliable project monitoring and stronger evidence for compliance and quality assurance. Over time, the robots' capabilities will expand to site logistics, dexterous manipulation, and multi-robot coordination as FieldAI's general-purpose systems grow more capable.
- 2
Brex builds AI agent control layer at network level, not in rules
Brex created CrabTrap, an open-source HTTP/HTTPS proxy that intercepts all network traffic from AI agents, examines policy rules, and uses an LLM-as-a-judge to approve or deny requests. The company found that traditional guardrails could not contain what agents were doing with real credentials like API keys and OAuth tokens. Brex's approach addresses a gap in how AI agents are currently governed—frameworks like OpenClaw enable agents to act, but lack enterprise-scale safeguards. By enforcing policy at the network layer rather than in the agent's code, organizations can audit and control agent behavior in real time, even when agents have genuine credentials to systems that matter.
Brex CEO Pedro Franceschi frames this as a shift in how IT leaders should think about agent governance: moving from SDK-level permissions and model guardrails to centralized network control. How widely CrabTrap is adopted, and whether other enterprises adopt similar network-layer enforcement, will signal whether this architectural approach becomes standard practice.
- 3
Smartsheet launches AI agent interface on AWS; cuts token use 35–47%
Smartsheet built a remote Model Context Protocol (MCP) server on AWS that gives AI clients like Amazon Quick and Claude Desktop direct access to Smartsheet's data and APIs. The server uses AWS Fargate, Amazon Kinesis, Amazon Neptune, and other services to let AI agents autonomously manage tasks, create sheets, and update projects—compressing workflows that took weeks into days or hours. Enterprise teams deploying AI agents need structured, secure access to internal data. Smartsheet's MCP server solves this by running behind the same security layer (AWS WAF, AWS Shield, OAuth2) as production APIs, with built-in governance—administrators can restrict agents to read-only or allow full write access per organization. Since launch, Smartsheet has saved over 3 billion tokens through AI-optimized serialization and progressive disclosure, directly reducing LLM inference costs.
Smartsheet optimizes token consumption at three levels—progressive disclosure that caps response size, strongly typed schemas to prevent hallucinated parameters, and a proprietary serialization format that reduces token count by 35–47 percent on data-heavy responses. Deployments roll to the smallest region first, validated by canary tests running every 15 minutes against the live environment.
- 4
GPT-5-6 Sol, Kimi K3 launched; AI regulation calls mount
Multiple AI models launched this week, including GPT-5-6 Sol, Kimi K3 (rolling out now), Muse Spark 1.1 from Meta, and Inkling from Thinking Machines. Demis Hassabis issued a call for regulatory action, and a new open letter on AI regulation was published. An Opus 5 announcement is expected soon. The pace of model releases signals ongoing competition among AI developers to deploy new capabilities. The simultaneous push for regulatory frameworks suggests tension between rapid deployment and calls for governance—a pattern the source describes as part of a larger weekly news cycle that now requires splitting into multiple parts.
Kimi K3 will be covered in detail next week. An Opus 5 announcement is likely coming soon. The regulatory calls by Hassabis and the open letter indicate growing pressure for policy action alongside continued model development.
- 5
EU AI Act OpenRAG: 933 legally structured chunks in SQLite
A developer has released EU AI Act OpenRAG, a downloadable corpus of EU Regulation 2024/1689 organized into 933 chunks structured by the law's legal framework (articles, recitals, definitions, annexes) rather than character windows, paired with 1024-dimensional BGE-M3 embeddings in a single SQLite database file. The structured approach improves retrieval accuracy for legal AI tasks — scenario article recall@20 reached 0.541 versus 0.449 for a baseline, and QA article hit@10 reached 0.927 versus 0.898 — making it more reliable for RAG (retrieval-augmented generation) and legal-NLP experiments on the EU's AI rulebook.
The corpus includes exact EUR-Lex links, Article 113 application-date metadata, and deliberately narrow derived labels with ambiguous cases marked NULL, designed to let researchers distinguish direct textual classification from broader regulatory-regime association.
- 6
Xi launches China's AI governance body at Shanghai conference
President Xi Jinping unveiled a package of measures at the opening of the 2026 World Artificial Intelligence Conference (WAIC) in Shanghai aimed at institutionalizing China's bid to shape global AI governance. China is moving beyond ad-hoc statements to establish formal structures for influencing how AI is governed worldwide, positioning itself as a central voice in setting international AI standards and norms.
The specific mechanisms and member countries of the newly unveiled cooperation body, and whether Western nations engage with or counter China's governance framework.
What to Watch
Watch for whether CrabTrap and similar network-layer enforcement approaches become standard practice across enterprises, signaling a fundamental shift in how organizations govern AI agents beyond traditional guardrails. Simultaneously, monitor the upcoming Opus 5 announcement and the regulatory pressure from figures like Hassabis alongside continued advances in models like Kimi K3—these parallel tracks of technical capability and policy action will reveal whether the industry and governments can coordinate on AI governance standards.
Sources
- McLaren Construction to deploy autonomous robots at scale in partnership with FieldAI
- Brex built its AI agent policy by watching what agents actually do, not by writing rules first
- How Smartsheet built a remote MCP server on AWS
- AI #177 Part 1: Tip of the Iceberg
- EU AI Act OpenRAG: 933 legally structured chunks and BGE-M3 embeddings in one SQLite file [P]
- China moves to institutionalize its vision for global AI governance as Xi launches cooperation body
- Aurora Group expands enterprise AI services with 12 workplace use cases
- New York governor says she’s using AI to analyze ‘every single rule’ in the state
- Microsoft (MSFT) Faces Browser Scrutiny As New York Clouds Data Center Expansion
- Japan revises AI policy guidelines to bolster cybersecurity
Share this with a friend
Send today's roundup to anyone who wants to keep up.
Get daily AI news free with AIToday
200+ AI sources, summarized in 1 minute. Email / LINE / Slack.
Sign up free