Russia's AI law strips copyright barriers

Russia's AI law strips copyright barriers — and bets on legal speed to outrun a chip shortage

The Federation Council approved a framework law on July 18 that grants domestic AI developers a broad copyright safe harbor for training neural networks, while reserving to the state the power to mandate "sovereign" models for public administration. The law takes effect on September 1, 2026 — and it is Moscow's answer to a hardware deficit no regulation can fix.

The Russian Federation Council on July 18, 2026 approved a framework law that will take effect September 1 — and it does something no Western jurisdiction has dared. The law declares that using copyrighted works for extraction, comparison, classification, and pattern detection with AI does not constitute copyright infringement, provided the developer obtained "lawful access" and the material is available for analysis without technical restrictions, according to a WAM report via TASS. AI developers may also use copyrighted works to train neural networks under the same conditions.

This is not an oversight. It is a calculated bet — one that says legal permissiveness can substitute for the silicon sanctions have denied.

Only if we have domestic models can we ensure technological progress, as well as maintain security, defence capability, and — which is vital — stay on the cutting edge in scientific and engineering fields.

That was Vladimir Putin in February 2026, addressing the Presidential Commission on AI he had just created, according to the Kremlin transcript. The law now passed is the hard legislative instrument behind that rhetoric.

A safe harbor the EU won't match

The copyright provision is the law's most consequential departure from every major regulatory effort elsewhere.

In the European Union, the AI Act — enforced since summer 2025 — requires general-purpose AI providers to comply with the text-and-data-mining exception in the 2019 Copyright Directive. That exception, under Article 4(3), explicitly allows rightsholders to opt out of having their works mined for commercial AI training, and providers must respect that opt-out, as codified in Recital 105 of the AI Act. The United States has no statutory text-and-data-mining exception at all: training on copyrighted works is fought case-by-case under the fair-use doctrine, with courts still split.

Russia's new law eliminates that uncertainty — for domestic firms, at least. There is no opt-out. There is no fair-use balancing test. If a developer obtained lawful access — a phrase whose boundaries will be drawn by Russian courts — the training is legal.

The contrast is stark enough to amount to an industrial-policy choice. According to a RAND Corporation analysis, the EU framework, while permissive in requiring opt-out, still imposes "significant transparency obligations" and "limits on commercial use." Russia's law imposes no such limits — and the transparency it requires is directed outward, at service providers who must inform users who owns copyright on AI-generated content, not at the developers themselves.

The hardware reality: 65 nanometers and a prayer

None of this legal scaffolding matters if there is no silicon to run the models on.

Western export controls have gutted Russia's access to advanced GPUs. Domestic chip fabrication operates at 65-nanometer technology, roughly fifteen years behind the cutting edge, according to the Carnegie Endowment. The country's two main processor designers — MCST and Baikal Electronics — lost access to TSMC's foundries after the February 2022 invasion, and Baikal's ARM architecture license was suspended after UK sanctions. Both firms are now effectively frozen at pre-war design nodes.

Imports have pivoted almost entirely to China. Over 80% of Russia's semiconductor imports now come from China, according to an AEI study by Chris Miller. But those chips are older, lower-grade, and arrive with reported defect rates as high as 40%, according to the U.S. Treasury, while Russia has paid roughly double the pre-war price for semiconductors, the same AEI analysis found.

This is the context for Putin's instruction in January 2026 — one month before the AI Commission was created — to have the government, Rosatom, and Rosseti draft a data-center infrastructure plan through 2030, with projections to 2036, tied to the rollout of nuclear, coal, and hydroelectric generating capacity, according to the Kremlin's published order. The same document called for "training multimodal AI models based on information about Russian culture, history, traditions and values."

The new law turns that executive wish list into statutory text. It grants the president authority to approve the National AI Development Strategy and explicitly empowers the government to specify cases in which only sovereign or national large foundational AI models may be used.

The target customer, in other words, is already locked in: the Russian state.

WAICO and the geopolitics of "technological sovereignty"

Russia's legislative sprint cannot be read in isolation. On July 16 — two days before the Federation Council vote — China formally launched the World Artificial Intelligence Cooperation Organisation (WAICO) in Shanghai, a 29-nation body that includes Russia, Indonesia, Brazil, South Africa, and Pakistan, as Al Jazeera reported.

President Xi Jinping, speaking at the World AI Conference on July 17, declared that "AI development should not be a solo performance by a single country, but a symphony of international cooperation," and called for "jointly opposing overstretching the national security concept in the field of AI," in a clear swipe at U.S. export controls.

Russia's framework law operationalises that message inside its own borders. The law enshrines "technological sovereignty" as a regulatory principle and gives the government a statutory tool to enforce it — the power to mandate domestic AI models. That mandate, combined with the copyright safe harbor, means that Russian firms can now train on vast corpuses of scraped or lawfully accessed global content, build models, and have those models guaranteed a government market, insulated from foreign competition by procurement rules and sanctions.

The Kremlin has been openly building toward this. At the Eurasian Economic Forum in May 2026, Putin proposed a "high-level meeting on artificial intelligence next year" to discuss "sovereign artificial intelligence models" and an "interconnected infrastructure of computing power and energy facilities," according to the Kremlin transcript. The data-center plan tied to nuclear power plants — Russia's one genuine competitive advantage in energy-intensive AI compute — is the physical leg of that strategy.

Who wins, who loses

Sberbank and Yandex are the immediate corporate beneficiaries. Both already operate large language models — YandexGPT and GigaChat — and both have been named in Putin's instructions as key partners. The copyright safe harbor removes the risk of domestic infringement litigation, while the sovereign-mandate provision directs state procurement toward their products.

The Russian state security apparatus gains a statutory framework for embedded, domestically controlled AI across critical infrastructure. This follows the pattern of earlier digital legislation — the sovereign-internet law of 2019, the Yarovaya data-retention law of 2016 — which progressively closed Russian cyberspace under the banner of sovereignty, as the Council on Foreign Relations documented. The AI framework law is the logical next instrument in that sequence.

Foreign rights-holders lose. The EU's opt-out mechanism has no Russian analogue. A novel published in France can be scraped and fed into a Russian neural network, and the author has no legal recourse under Russian law — so long as the work was lawfully accessed and not behind a paywall or technical restriction. The law does not define what constitutes such a restriction with precision; that ambiguity itself is a form of permission.

Chinese chip suppliers hold the swing position. Russia's models will remain constrained by whatever GPU equivalents China can provide without triggering secondary U.S. sanctions. If Beijing chooses to prioritise WAICO solidarity over compliance with Western export controls, Russia's legal advantage becomes operationally meaningful. If not, the law remains a well-drafted document without the silicon to execute it.

Diplomat View

The Russian Federation Council's AI law is a structural wager: that the state can accelerate domestic AI development by removing legal friction faster than sanctions can degrade the hardware base required to run it.

The logic is coherent. The copyright safe harbor offers Russian firms a training-data advantage that their U.S. and EU competitors — mired in litigation and opt-out compliance — do not enjoy. The sovereign-mandate provision guarantees a state buyer. The data-center plan, if executed, could give Russia a competitively priced energy backbone for AI compute.

But the wager has a hard ceiling. Without access to GPU-class chips at scale, Russian frontier models will lag their Western and Chinese counterparts regardless of how much copyrighted text or code they are free to ingest. AEI data shows sanctions have already forced Russia to pay double for semiconductors and to rely on Chinese intermediates with high defect rates. The Kremlin's pledged $38 billion for microelectronics by 2030 is a fraction of the U.S. CHIPS Act's $280 billion — and the talent pool is shrinking, not growing.

The critical variable is whether the U.S. imposes secondary sanctions on Chinese firms found re-exporting advanced chips to Russian AI labs. The AEI study demonstrates that new sanctions designations can cut shipments to targeted Russian firms by roughly 40% within two months. If Washington escalates enforcement in the wake of the WAICO launch and Russia's legislative push, the legal advantage becomes largely theoretical.

What to watch:

• Presidential signature and September 1, 2026 implementation. The law is expected to be signed without changes. Watch for the first government decrees specifying which sectors must use "sovereign" models — this will reveal the Kremlin's AI procurement priorities.
• U.S. Treasury/BIS sanctions designations in Q3–Q4 2026. Any expansion of the Entity List targeting Chinese chip distributors supplying Russia will directly constrain the hardware base this law is designed to exploit.
• The next WAICO ministerial meeting. The 29-nation coalition is the diplomatic vehicle for the "sovereign AI" model — its institutional heft, or lack thereof, will signal whether Russia's legislative bet has an international market or remains a domestic play.

The bottom line

Russia's framework AI law is not primarily about regulating AI. It is about constructing a legal environment in which domestic firms can train models on whatever data they can lawfully reach — without fear of litigation, without an opt-out for rightsholders, and with a guaranteed government buyer at the end of the pipeline. The hardware deficit is real and worsening, but the Kremlin has judged that legal speed can bridge the gap until — or unless — sanctions tighten further. The law takes effect on September 1. The chips that will determine its actual impact depend on decisions made in Washington and Beijing, not Moscow.