AI’s power struggle heats up: EU readies record Google fine as model theft claims spread

On May 22, 2026, multiple outlets highlighted how AI is colliding with politics, competition law, and IP enforcement. One report says the EU’s Digital Markets Act (DMA) process against Google is reportedly close to concluding, with the bloc preparing what could be a record fine. Separate coverage frames Google’s latest technology as especially challenging for OpenAI, which has led in consumer AI, suggesting a fast-moving competitive race. Meanwhile, NRC reports that OpenAI, Anthropic, and Google are accusing each other of stealing expensive AI models, warning that chatbots can reveal how they work when prompted correctly. Strategically, the cluster points to a shift from “AI as a product cycle” to “AI as a geopolitical instrument,” where regulatory leverage and IP control become tools of market power. The EU’s willingness to impose a potentially record DMA penalty signals that regulators want to constrain gatekeeper behavior in AI-adjacent ecosystems, not just traditional search and ads. At the same time, the model-theft allegations imply that competitive advantage may be increasingly extracted through reverse-engineering, data access, or imitation—raising the risk of a tit-for-tat enforcement spiral. Companies and governments both benefit from stronger enforcement narratives, but consumers and smaller developers could lose if compliance costs rise or if litigation chills model sharing and interoperability. Market and economic implications are likely to concentrate in AI software platforms, cloud inference capacity, and ad-tech/search ecosystems. A record DMA fine risk can pressure Google-related equities and sentiment around EU regulatory exposure, while also boosting demand for compliance tooling and legal services across the sector. The “utilities sector” angle in one article suggests AI-driven optimization is moving into critical infrastructure-adjacent domains, where political scrutiny can translate into procurement delays and higher risk premia for vendors. If model theft claims lead to tighter access controls, the near-term effect could be higher costs for training and deployment, with knock-on impacts for the competitive positioning of OpenAI, Anthropic, and Google in consumer AI. What to watch next is whether the EU formally issues the DMA decision and the size/timing of any fine, since that will set a benchmark for future enforcement. In parallel, monitor court filings, technical disclosures, and any coordinated industry moves on model provenance, watermarking, and access controls, because the allegations described could escalate quickly. A key trigger point is whether regulators treat model extraction and imitation as a competition or IP violation under existing frameworks, which would broaden enforcement beyond advertising/search. Over the coming weeks, investors should track signals of procurement or partnership slowdowns in AI-enabled critical services, alongside any public statements by OpenAI, Anthropic, and Google responding to the theft claims and the EU’s process.

Geopolitical Implications

• 01

  AI governance is becoming a cross-border power lever: the EU’s enforcement posture can reshape global AI market structure and bargaining power.

• 02

  IP and model-provenance disputes may accelerate fragmentation of AI ecosystems, reducing interoperability and increasing compliance-driven barriers to entry.

• 03

  Competitive dynamics between US AI leaders (OpenAI/Anthropic/Google) are increasingly intertwined with regulatory outcomes, turning corporate strategy into quasi-geopolitical positioning.

• 04

  Critical-infrastructure adjacency (utilities) raises the stakes of AI oversight, where political scrutiny can translate into procurement and deployment constraints.

Key Signals

• —Formal EU DMA decision date and the fine amount (or settlement signals) for Google
• —Court filings or regulator statements on whether model extraction/stealing is treated as competition harm or IP infringement
• —Industry adoption of watermarking, provenance standards, and access controls to mitigate extraction
• —Public responses from OpenAI, Anthropic, and Google on the theft allegations and any changes to model release policies