Rio Tinto's legal team (OTCMKTS: RTNTF) just demonstrated something that should get every knowledge management practice's attention: they've successfully integrated ChatGPT's Deep Research capability with their iManage system, enabling 30-minute deep research sessions across millions of documents with full citations.^[1]

The First Real-World Deep Research Implementation?

Christopher de Waas, Rio Tinto's Digital Transformation Lead, reported that their legal team can now "query decades of institutional knowledge, at scale" using ChatGPT Enterprise connected to over 4.5 million documents in iManage.^[1] This isn't a pilot or proof-of-concept — it's a working system that generates reports up to 15,000 words with proper citations. The implementation has already achieved 80% user engagement within four months, suggesting the AI legal technology is solving real problems rather than creating new ones.^[2]

What makes this particularly relevant for contract lawyers at law firms is the scale and complexity Rio Tinto handles. As a global mining company operating in 35 countries with 150+ years of documentation, their legal challenges mirror those faced by major law firms: massive document volumes, complex historical precedents, and the need to surface relevant information quickly across jurisdictions.^[3]

The Economics Just Shifted Dramatically

The timing of Rio Tinto's implementation coincides with OpenAI's 80% price reduction for their o3 reasoning model — from $40 per million output tokens to just $8.^[4] This pricing change moves sophisticated reasoning capabilities from expensive experimentation to routine deployment territory for AI legal assistants.

Box CEO Aaron Levie captured the strategic implication of the o3 price reduction: rather than building AI features that are "economically reasonable today," companies should now "build for where AI is going" because inference costs are dropping by orders of magnitude.^[5] For legal teams, this means document analysis workflows that seemed cost-prohibitive last quarter are now affordable for regular use.

Understanding AI Limitations vs. Measurement Artifacts

A separate development this week provides crucial context for evaluating AI tools. Researchers challenged Apple's widely-cited study claiming that large language models hit hard reasoning limits, showing that the apparent failures were actually due to token constraints, not cognitive limitations.^[6] When researchers allowed the same models to provide compressed responses rather than step-by-step outputs, performance dramatically improved.

This finding matters for legal practitioners evaluating AI-powered legal tools. What appears to be a reasoning failure might actually be a formatting or context window constraint. The lesson: when AI tools seem to hit walls, consider whether the limitation is in the underlying capability or the way the task is structured.

What This Means for Your Practice This Month

The Rio Tinto implementation offers a template that's immediately actionable. Their approach focused on three practical elements:

Integration with existing workflows: Rather than replacing iManage, they built Deep Research on top of it, preserving existing access controls and user permissions.^[1]

Realistic expectations: They acknowledge this isn't a replacement for legal judgment but a tool for "improving access to millions of docs" and enabling "complex searches far beyond what users could realistically handle manually."^[1]

Incremental deployment: They're starting with document search and synthesis before expanding to other AI capabilities.

For firms with large document repositories, the combination of Deep Research capabilities and dramatically reduced reasoning model costs creates an immediate opportunity. The technology stack Rio Tinto used—ChatGPT Enterprise with Model Context Protocol (MCP) integration—is available now, not in development.

The Broader Context

While industry observers debate long-term AI impacts on legal jobs, the Rio Tinto example shows how AI is actually being deployed: as a powerful search and synthesis tool that amplifies human expertise rather than replacing it.

The key insight from this week's developments isn't about AI's eventual potential—it's about recognizing that sophisticated capabilities are already accessible at affordable pricing for organizations ready to implement them thoughtfully.

If you're interested in exploring how these AI legal automation capabilities could transform your practice, our team specializes in helping law firms implement similar solutions. Contact us to learn more about bringing enterprise-grade AI legal assistants to your organization.

References:

^[1] LinkedIn post by Christopher de Waas, December 11, 2024. https://www.linkedin.com/posts/cdewaas_chatgpt-legaltech-knowledgemanagement-ugcPost-7338182904826470402-M0Lg/

^[2] iManage press release, "Rio Tinto Selects iManage to Support Legal Transformation Across Global Operations," June 12, 2025. https://imanage.com/resources/resource-center/news/rio-tinto-selects-imanage-to-support-legal-transformation-across-global-operations/

^[3] Rio Tinto company profile. https://www.riotinto.com/en/about/our-business

^[4] OpenAI Community announcement, "O3 is 80% cheaper and introducing o3-pro," June 10, 2024. https://community.openai.com/t/o3-is-80-cheaper-and-introducing-o3-pro/1284925

^[5] Aaron Levie Twitter post, June 10, 2024. https://x.com/levie/status/1932617875358896598

^[6] Rohan Paul Twitter thread on Apple reasoning study critique, December 9, 2024. https://x.com/rohanpaul_ai/status/1933296859730301353