Boundless Offers AI with Denial Response Removed under Review System for ZK Proofs

Key Points of This Article

• The first version is the DeepSeek V4.1 Flash denial suppression version
• In July, approximately 4,000 GPUs for ZK will be opened for AI inference

Providing Denial Suppression Model under Review System

Boundless, which operates a distributed GPU network for zero-knowledge proofs (ZK proofs), announced on the 24th that it has started accepting applications for its AI model provision program "Defensive Access" aimed at security researchers. Only teams that pass the review will be able to use the modified AI model that suppresses denial responses.

As the first offering, a denial suppression version of the open-weight model "DeepSeek V4.1 Flash 1017" from China’s DeepSeek will be provided. The intended use is limited to security research and red team exercises (simulated attacks by role-playing attackers) and penetration testing, with prior authorization required. Applicants must indicate the system to be tested and the intended use, which the company will review on a case-by-case basis.

An open-weight model refers to an AI model with publicly available learned weights that anyone can run or modify. The version provided by the company is called "Abliterated," which has modifications to suppress denial responses. The company also states that reducing denials does not necessarily improve the model's accuracy and that verification of outputs is necessary.

There are other providers offering open-weight Abliterated models under a review system. The inference service Inference.net provides the same denial suppression version of DeepSeek V4.1 Flash 1017, but only to approved teams.

Aiming to Support the Defensive Side and Risks of Misuse

Boundless cites the reason for its provision as the issue that general AI models may deny legitimate test requests that resemble attack methods. The company’s team stated in the release, "The safety restrictions of traditional models can backfire and become a hindrance to those trying to protect the system."

Models with suppressed denial responses could be used to generate malicious code if they fall into the hands of attackers. The company has not disclosed specific criteria for the review or how it will monitor usage after approval.

In terms of performance, the company refers to initial benchmarks published by AI evaluation firm Vals AI on the 20th. When tasked with multi-layered cybersecurity tasks simulating attacks on a mock corporate network, the standard version of the model reportedly outperformed OpenAI's "GPT-6 Sol." This result pertains to the standard version before modifications and does not evaluate the denial suppression version itself.

From ZK Proof Network to AI Inference

Boundless originally built a GPU network to generate ZK proofs for the cryptocurrency industry. On July 14, it announced that approximately 4,000 GPUs would be opened for AI inference, operating as a managed inference service (where the company handles operations). The operation of the ZK proof network continues in parallel.

The company plans to give its native token "ZKC" a role in the AI network, indicating plans to require ZKC staking from AI inference operators. The announcement of Defensive Access does not touch on its relationship with ZKC.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.