When Washington Declares War on AI Theft
The White House turned up the heat this week. The U.S. administration’s technology policy office officially condemned what it calls “industrial-scale” AI theft orchestrated by foreign entities, with a pointed finger at Chinese firms. The method? No spectacular hacking or Hollywood-style espionage. Bad actors would operate through proxy accounts—in other words, fake profiles or intermediaries—and resort to jailbreaking, a technique to bypass safeguards built into AI models in order to extract normally restricted capabilities.
The idea in practice is to extract trade secrets from a major American language model, then reuse them to train a competing model at a fraction of the cost. This is called model distillation: instead of spending billions on infrastructure and data, you generously borrow from others’ work. A sort of illegal digital photocopy, and particularly hard to detect.
Washington is now promising concrete measures to counter these practices, though details remain vague for now. One thing is certain: the Sino-American tech war has opened a new front, and this time it’s algorithmic.
Tencent Quietly Releases Hy3, Yet Makes Waves
As if to prove Washington’s nerves are frayed, Tencent just released a preliminary version of its new AI model called Hy3 as open-source. To put it mildly, the Chinese tech giant isn’t hyping it up quite enough yet.
Developed in less than three months—a timeline that would make more than one Western lab blush—Hy3 shows impressive performance across several key tasks: assisted coding, logical reasoning, and information retrieval. In other words, exactly the areas where models like GPT-4 or Claude have stood out for months.
What makes Hy3 particularly interesting is its performance-to-resource ratio. The model reportedly outperforms similarly-sized competitors, meaning less computing power is needed for equivalent results. In a context where AI training costs run into the hundreds of millions, efficiency has become the new battleground.
Tencent chose to make Hy3 open-source, a strategy now standard in the industry to gain adoption and credibility. The irony? While Washington accuses Chinese entities of stealing American technology, Beijing is opening its own code to the world. AI geopolitics is rarely straightforward.
Tether Strikes Hard: $344 Million in USDT Frozen on Tron
Shifting gears now, but staying in the realm of surveillance and control. Stablecoin issuer Tether announced it froze $344 million in USDT on the Tron blockchain, in coordination with OFAC—the U.S. Office of Foreign Assets Control—and American judicial authorities.
This ranks among the largest freezes ever executed in stablecoin history. To recap: freezing USDT means the affected wallets become completely unable to move or spend their funds. The blockchain keeps running, the tokens technically still exist, but they’re rendered unusable—like a frozen bank account, but on a decentralized chain.
The targeted addresses had been flagged by U.S. authorities, though specific details of suspected activities haven’t been made public yet. Such actions remind us that, contrary to a still-common misconception, cryptocurrencies aren’t synonymous with absolute anonymity or total impunity. Tether, as USDT’s centralized issuer, technically retains the ability to act on funds—making it both a compliance tool and a centralization point that critics are quick to highlight.
This operation reflects a broader trend: American authorities are strengthening cooperation with crypto industry players to trace and block suspicious financial flows, with growing effectiveness.
Taking Stock: Surveillance, Competition, and Control
These three seemingly separate stories actually tell the same tale: states and major institutions trying to regain control over technologies originally designed to escape centralized oversight.
Open-source AI spreads at a pace even its creators struggle to manage. Stablecoins circulate on supposedly decentralized blockchains, yet remain subject to central issuers’ decisions. And the tech race between superpowers accelerates, blurring lines between legitimate innovation and contested appropriation.
In this landscape, one certainty emerges: the next decade will be as much about regulation as innovation. The real question is whether rules will follow technology… or whether technology will ultimately circumvent the rules.
