DeepSeek sparks some soul searching
The upheaval caused by the introduction of DeepSeek's open-source R1 model forces a rethink on AI priorities
It has been impossible to ignore the launch of DeepSeek, the Chinese startup's AI assistant, developed at a fraction of the cost and time of its Western counterparts. Its emergence has forced a complete re-evaluation of the AI industry's economics and competitive dynamics, challenging assumptions about the resources needed to produce high-performance AI systems.
The initial response to DeepSeek's debut was immediate and dramatic. The Nasdaq 100 index fell approximately 3%, with Nvidia's shares alone plummeting nearly 18%. These movements reflected investor anxiety over DeepSeek's potential to disrupt the market.
Its success directly challenges the dominance of firms that rely on capital-intensive infrastructure, demonstrating that cost-effective alternatives can compete with—and perhaps surpass—Western models in performance. The implications for industries relying on AI are profound, as the race to innovate will increasingly prioritise efficiency as much as scale. (Click here for a visualisation of the advancements of Chinese LLMs in the global AI race.)
But its arrival heightened my concerns about the broader risks associated with large language models (LLMs)., which focus on three key areas: intellectual property ownership, data privacy and security, and legal jurisdiction.
To explore these, I asked ChatGPT and DeepSeek to clarify their policies, uncovering distinct approaches that reveal both opportunities and risks.
Ownership remains one of the most contentious issues. Both platforms agree that users retain ownership of their input data, but the treatment of generated outputs varies. DeepSeek employs a shared ownership model: while users own their inputs and outputs, the platform reserves rights to use this data for improving its models. Additionally, its proprietary algorithms remain firmly in its control.
By contrast, ChatGPT enables users to claim exclusive ownership of outputs, provided they do not violate laws or ethical standards. However, OpenAI reserves the right to use user data for training unless explicitly instructed otherwise.
For organisations dealing with sensitive or proprietary information, these distinctions require careful scrutiny. Shared ownership or vague opt-out clauses can pose risks to intellectual property, particularly when innovation is key to competitive advantage.
Data privacy presents another critical divergence. DeepSeek’s policies emphasise compliance with regulations such as the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA), anonymising user data where necessary.
However, businesses integrating DeepSeek into customer-facing services must implement additional safeguards to ensure secure handling of sensitive information because the global regulatory landscape for AI is constantly evolving.
ChatGPT takes a slightly different approach, claiming not to use data submitted via its API for training unless users explicitly opt in. While this might seem more secure, vulnerabilities arise when ChatGPT is embedded in third-party applications. Here, data handling relies on the developer’s policies, making thorough evaluation of external integrations essential.
Legal jurisdiction compounds these challenges. DeepSeek operates under Chinese law and typically requires arbitration to resolve disputes. ChatGPT’s terms, governed by Californian law, rely on arbitration administered by the American Arbitration Association.
While arbitration can provide a more efficient resolution process, it is crucial to examine the specific terms of arbitration agreements to ensure they align with an organisation's operational and legal needs. For any global operation, fully understanding jurisdictional implications is essential before committing to either platform.
While both platforms share certain principles—compliance with privacy regulations, user ownership of outputs, and reliance on arbitration—their differences carry significant consequences for organisations. The nuances in jurisdiction, data use policies, and third-party integrations underscore the need for a thorough risk assessment before adoption.
Organisations must take a strategic and informed approach to mitigate these risks:
Intellectual property must be secured through clear agreements that grant exclusive rights to AI-generated outputs. Reviewing and negotiating terms of service is essential to avoid inadvertent data usage or ambiguity over ownership.
Safeguarding sensitive data requires rigorous evaluation of platforms and their ecosystem integrations. Strong encryption protocols, robust data-sharing agreements, and internal compliance audits should be standard practice.
Legal preparedness involves understanding jurisdictional requirements and anticipating the complexities of arbitration clauses. Organisations should seek expert legal counsel to ensure service agreements align with their risk tolerance and global operational footprint. Special attention must be paid to the terms of arbitration agreements to confirm their compatibility with an organisation’s dispute resolution preferences.
These risks are not abstract concerns but critical issues that can directly impact an organisation’s strategy and reputation. Platforms like DeepSeek and ChatGPT offer undeniable opportunities, but their adoption demands vigilance, transparency, and a proactive stance.
By asking the right questions now and addressing these challenges head-on, businesses can build a foundation for sustainable and compliant AI integration, prepared to navigate the evolving landscape of artificial intelligence.
You can find OpenAI’s Terms of Use here and those of DeepSeek’s here.
(A quick word on how I put this blog together. Firstly, I really struggled to create a DeepSeek account. Its popularity means that they are battling to cope with demand. When I eventually did, I asked my three questions of each LLM. Then I asked each to write a blog on it before combining them, rewriting and running it through Gemini for fact-checking. That process itself was fascinating. The prompt for the original answer was: “I want you to write an analysis of who owns information or products in ChatGPT, particularly when it is used in an agent, how private data is, and where legal disputes are domained.”