#329: The Treasury Department’s Crackdown On Tornado Cash Sets A Problematic Precedent For Cryptocurrency Privacy Controls, & More

Last week, the Treasury Department’s Office of Foreign Assets Control (OFAC) sanctioned Tornado Cash, a token mixing service that allows for privacy-preserving transactions on the Ethereum blockchain. OFAC noted that criminal organizations like North Korea’s Lazarus Group have leveraged the service for money laundering purposes. As a result, it added Tornado Cash’s website and 44 associated Ethereum addresses to its Specially Designated Nationals and Blocked Persons List (SDN). As required by US law, node operators Infura and Alchemy blocked access to Tornado Cash while GitHub took down the code repository, blocking the accounts of its contributors, and Circle froze the USDC in Tornado Cash wallets.

Importantly, Tornado Cash is a smart contract––a piece of open-source software––for crypto users seeking privacy. In other words, for the first time, OFAC has sanctioned code as the “throat to choke” in addressing privacy-related concerns, a decision with massive legal and financial implications for developers and downstream recipients of tokens that have touched the Tornado Cash service.

On-chain derivatives exchange, dYdX, blocked addresses that had touched Tornado Cash, shocking users who had no idea that their ETH had anything to do with it. Meanwhile, Dutch authorities arrested Alexey Pertsev, a Tornado Cash developer––and another “throat to choke”––in Amsterdam, claiming that his software contribution to this open-source code facilitated money laundering.

Sanctioning Tornado Cash to disrupt foreign hacking groups, OFAC seems to have used a blunt instrument to set a dangerous precedent that could limit privacy rights on the internet. Others in the crypto community agree. In our view, the sanctions underscore the importance of deep decentralization to preserve crypto’s independence in the long term.

The COVID-19 pandemic turbocharged the adoption of digital healthcare services. Digital therapeutics (DTx) are FDA-approved software products designed to prevent, treat, and/or manage medical conditions. The pandemic incentivized an expansion in virtual healthcare infrastructure that has not yet but could galvanize the adoption of DTx.

Historically, many DTx have received FDA-approval for behavioral and other chronic conditions, like anxiety and opioid use disorders, that involve cognitive behavioral therapy (CBT). Because some DTx analyze the patient’s data and require doctor prescriptions, large pharmaceutical companies like Novartis and Sanofi have agreed to co-commercialize them, sometimes in combination with drugs.

Despite those advances, DTx have not found broad-based success. After the FDA approves a DTx, the US Centers for Medicare and Medicaid Services (CMS) must assess its reasonableness and necessity before granting it reimbursement. In the absence of broad-based coverage, medical interventions struggle to survive. According to our research, until recently no commercial or public payer in the US has granted DTx far-reaching coverage policy. In April 2022, however, CMS issued a provisional billing code for prescription digital CBT, signaling its willingness to reimburse DTx.

According to our research, DTx could be much faster and less expensive to develop than standard pharmaceutical drugs. While not likely to be effective for some diseases, DTx could offer profound treatments for others. If CMS continues to support this new benefit category with coding guidelines, the DTx uptake could be transformational for many behavioral, neurological, and chronic conditions.

Although the results are mind-blowing, an AI model that will be able to edit a video realistically should come as no surprise. Like still-image AI systems, the new DALL·E 2 model enables video with text-based commands. The model can create videos of a tennis player playing on the precipice of a volcano or in the middle of the desert.

Conceptually, video should follow advances in still imagery. Given ARK’s forecasts for steep declines in AI training costs and ramps in hardware investment, we can estimate how much video costs might lag behind still-image costs. As discussed in ARK’s Big Ideas 2022, we expect AI training costs to drop 2.5x per year through 2030 and AI hardware spending to double annually through 2025, at least. The combination suggests that the capability of AI systems will improve multiple thousand-fold over the next five years.¹

How much more difficult is generating a 30-second video compared to a still image? A TV advertisement has roughly 2,000x more pixels than a still image created by DALL·E 2. With that ratio as a proxy, we believe that 30-second AI videos will be as productive and provocative by 2026 as are AI still-image generation systems today.

The implications for advertising, media distribution, and entertainment are profound. Why spend $100,000 on a single television commercial targeting millions of people when, with the same budget, an advertiser will be able to create 10,000 different commercials, each tuned to a cluster of like-minded viewers at a moment in time? Stay tuned.

[1] ARK expects annual investments to increase by more than 20x over five years and the cost of AI training to decrease by almost 100-fold over the same time frame. This suggests a capability improvement exceeding 2,000x.