Venice AI’s VVV token briefly rose above $25 on Tuesday as a public dispute over private AI research chats pushed “privacy inference” back into focus. The episode centered on whether sensitive material entered into AI tools can later contribute, even indirectly, to model improvement—a question with growing relevance for researchers, developers, businesses, and automated trading systems that increasingly place proprietary work inside model workflows.
The price move coincided with remarks by Tristan Buckmaster, a mathematician at New York University, who said he and collaborators had submitted a complete project draft to Codex. Buckmaster later asked whether the model had accessed the group’s private conversations or used their content in training.
OpenAI said researchers or agents did not access particular user data to solve the mathematical problem. The company also said it could not completely exclude the possibility that de-identified data from private Codex chats had contributed to broader model improvement, while characterizing that scenario as unlikely.
The exchange did not establish that the draft was used in training or that any individual accessed it. It did expose a practical concern around the boundaries of private AI use: confidentiality protections can vary sharply depending on whether a service hides identity, deletes logs, encrypts requests, or processes data in protected hardware.
Venice sells privacy options across model providers
Venice AI, founded by cryptocurrency entrepreneur Erik Voorhees, operates a consumer chat product and an API for developers. It aggregates access to multiple AI models and markets privacy controls and fewer content restrictions as product features.
Its design offers several levels of protection rather than a single blanket privacy guarantee. Venice’s anonymous mode removes user identity from a request, although the upstream model provider can still view the request content. Zero-retention mode relies on provider commitments not to retain user prompts and outputs. Its TEE mode runs inference inside a trusted execution environment, or TEE, a protected hardware area intended to limit access to data while computation is taking place.
The company also offers end-to-end encryption in configurations where a request remains encrypted from the user’s device until it is decrypted within the protected computing environment. This architecture is designed to reduce what an infrastructure operator can inspect, though its protections depend on the model, provider, and mode selected by the user.
That distinction has become more commercially relevant as AI systems move beyond public-facing question-and-answer tasks. A casual prompt and an unpublished research paper present very different confidentiality risks. So do source code repositories, internal product plans, customer records, trading logic, and legal work. Users seeking protection need to determine where their request is decrypted and whether it is routed onward to another provider.
Revenue growth and token mechanisms support VVV interest
Banyan, which described itself as a Venice investing participant, said Venice’s annualized revenue increased from $14 million in January to more than $100 million in August. The figure offers one of the few public indicators of demand for Venice’s chat and API services, although the company’s full financial statements are not public.
VVV, issued on Coinbase’s Base network, is connected to API spending through a buyback-and-burn program. Venice allocates $5 for every $100 spent on API purchases to acquire and burn VVV, reducing the outstanding supply of the token. The arrangement ties token demand partly to paid usage, though the size and timing of market purchases will depend on API revenue.
Venice also reduced annual VVV emissions from 3 million tokens to 2.5 million on Sept. 1, with a further decline to 2 million scheduled for Oct. 1. Lower emissions reduce the number of newly issued tokens entering circulation, placing greater emphasis on whether product revenue and API usage can sustain demand.
A second utility channel comes through DIEM, an asset users can mint by locking staked VVV. Staking one DIEM produces a refreshed $1 daily allocation of Venice API credits. The design gives developers and autonomous software agents a way to hold an asset linked to recurring future model usage rather than simply purchasing API credits as needed.
DIEM’s target supply is due to rise from 38,000 to 40,000 on Sept. 14, completing a phased expansion that creates capacity for additional minting. That increase could support greater API-credit distribution, but it also gives the market a near-term supply variable to watch alongside VVV’s scheduled emission reductions.
NEAR extends the private-inference model
NEAR is also positioned around the same privacy-inference theme through an integration announced in March between Venice and NEAR AI. The arrangement allows Venice users to select verifiable private inference from NEAR AI and other providers, with Venice initiating the request and privacy-focused computation handled at the infrastructure layer.
NEAR AI Cloud uses TEEs to run models in an environment where plaintext is intended to remain inside a protected enclave during computation. Hardware attestations allow users to verify that a request entered the specified environment, offering a technical check on the claimed execution conditions.
The protection has clear limits when a request is sent outside that environment. Open-weight models can be deployed within the TEE setup, while requests routed through gateways to closed systems such as Claude, GPT, or Gemini must still be passed to their upstream providers. Those calls sit outside NEAR’s confidentiality boundary.
NEAR introduced another token-linked feature on July 30, allowing holders to convert staking yield into inference credits. The credit amount varies with the amount staked, NEAR’s market price, and the staking yield rate. Users can unstake to leave the arrangement, making the feature different from a permanent token burn or a fixed credit subscription.
NEAR’s agent infrastructure also includes NEAR Intents, an execution framework in which users submit a desired outcome—such as a cross-chain swap—and competing solvers seek to complete it. DeFiLlama reported that NEAR Intents generated about $9.32 million in total fees during the second quarter, with roughly $1.5 million retained by the protocol. NEAR uses retained revenue for market buybacks of its native token.
Bittensor links AI services to subnet competition
Bittensor takes a different route to AI-related token utility. Its network is organized into subnets, where miners supply specialized services including inference, storage, and prediction, while validators assess output quality and distribute rewards.
Chutes, a Bittensor subnet focused on model inference, reported about $1.37 million in second-quarter revenue from subscriptions, usage-based charges, and instance services. Its figures illustrate how Bittensor’s model separates AI markets by task rather than operating a single unified inference service.
Each Bittensor subnet has its own alpha token paired with TAO in liquidity pools. When users stake TAO into a subnet, it is converted into that subnet’s alpha token, making TAO the base asset used to direct stake across competing services. That structure gives participants exposure to specific subnet activity while retaining TAO as the network’s common economic layer.
TAO has a hard supply cap of 21 million tokens. Bittensor completed its first halving in December 2025, and current issuance stands at 0.5 TAO per block, or about 3,600 TAO per day.
The renewed attention around AI data handling places a sharper divide between services that promise privacy through policy and those that attempt to enforce it through technical architecture. For Venice, NEAR, and Bittensor-linked inference providers, sustained interest will depend less on the rhetoric around private AI than on whether developers choose their systems for workloads where confidentiality, verifiability, cost, and model quality must all be balanced.
Explore how AI copy trading leverages automated agents and data-driven strategies, complementing privacy-focused AI tokens like VVV.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
