AI is stretching patent law — and predictability is starting to fray
12th IP and Competition Forum
Key Takeaway: Artificial intelligence is forcing patent law to confront questions it was never designed to answer. Is training an AI system really the same as programming? Can a probabilistic, data-driven process satisfy doctrines built on predictability and objective technical problems? And if not, what must change to preserve incentives to innovate?
Those questions framed a practitioner-led panel at OxFirst’s 12th IP and Competition Forum, held in Oxford on January 13 & 14 2026. Panelists discussing patents and AI, revealed broad agreement on one point: existing patent principles are being stretched, but the system cannot afford to lose predictability.
Head of IP at Volvo Cars, Mr Graham, opened by challenging the core analogy underpinning many AI patent debates. Treating AI training as conventional programming, he suggested, only works up to a point. Over time, that analogy becomes strained, and new principles may be needed to reflect how AI systems actually function.
Mr Thomas, Senior Patent Attorney at Qualcomm, broadened the lens. AI, he said, should not be equated solely with consumer-facing tools such as large language models. Increasingly, it is embedded deep inside technical systems — communications networks, video coding and other engineering applications — where it tackles problems that do not admit clear, analytically defined solutions. AI shifts innovation from solving a known problem to discovering workable outcomes in environments characterised by scale, data and uncertainty.
That shift sits uneasily with patent law’s traditional assumptions. Patents are built around the idea of an objectively defined technical problem solved in a reproducible way. AI, by contrast, often produces heuristic solutions that cannot be fully explained or analytically derived. Thomas welcomed efforts by institutions such as the EPO to revisit how future inventions should be assessed, stressing that AI must remain patentable if innovation is to reach the market.
Predictability versus black boxes
Professor Ghafele, Managing Director at OxFirst, steered the discussion toward patent quality and applicant expectations. From an innovator’s perspective, Mr Thomas responded, predictability is paramount. Investment decisions depend on knowing whether protection is likely and enforceable. International landscaping studies already show diverging treatment of AI patents across jurisdictions, heightening the importance of a coherent European framework.
In practice, Mr Graham noted, AI is currently deployed less as a source of radical invention than as a productivity tool — doing more with fewer resources. Chair of Board of Appeal 3.5.05 of the EPO, Dr Bengi- Akyürek injected caution into claims of efficiency. AI outputs, he said, still require extensive human verification. Far from eliminating work, they often generate new layers of checking.
Mr Thomas pointed to a deeper limitation. Today’s AI systems are probabilistic and prone to hallucination. They do not yet deliver the level of certainty required for legal proceedings — and may never fully do so. That raises uncomfortable questions about how much uncertainty patent law can tolerate.
From an enforcement perspective, opacity cuts both ways. Dr Bengi- Akyürek observed that the black-box nature of AI makes infringement harder to detect, but also easier to design around. Mr Graham added that AI-based portfolio analysis tools could eventually change litigation dynamics. Large companies often avoid suing each other because the analytical burden is too high. If AI makes portfolio analysis reliable and cheap, long-standing strategic stand-offs may erode.
The next frontier
Looking ahead, Mr Thomas noted that despite enormous R&D investment, the number of clearly identifiable AI patents remains relatively modest — for now. That is likely to change as AI use cases multiply. Prof Ghafele pushed the panel to look further ahead, asking how AI will reshape SEPs and technologies such as 6G.
Mr Thomas said it is still too early to predict outcomes, but AI will be central to next-generation standards, potentially triggering paradigm shifts across industries. Mr Graham illustrated the stakes with autonomous driving. It remains elusive, but the scale of investment shows how commercially consequential AI patentability already is.
Institutional consistency loomed in the background. Prof Ghafele raised concerns about divergence between the EPO and the UPC. Mr Thomas expressed hope for convergence, though he acknowledged that the system is still young. Audience interventions suggested that claim construction differences exist, but not beyond the normal variation between legal bodies. Dr Bengi- Akyürek agreed that while claim construction is decisive, divergence may turn on how patent descriptions are used — as explanatory context or limiting constraints.
The discussion closed on a familiar theme. Greater harmonisation and predictability are urgently needed, Mr Thomas argued, because investment decisions are made at the start, not after years of litigation. Dr Bengi-Akyürek cautioned that harmonisation is difficult, but from the innovator’s perspective the central question remains simple: what protection will this invention receive, and where?
As Prof Ghafele concluded, in an AI-driven world the battleground will be inventive step — and whether patent law can adapt without losing the certainty innovation depends on.
Note: The views of the speakers do not necessarily reflect those of OxFirst, its affiliates, or employees. They also do not represent an official view of the companies the speakers work for. They are only the personal views of the speaker.