Trishool (SN23) Expanded Into Commercial AI Safety With Astroware

Trishool (SN23), a subnet focused on identifying weaknesses in artificial intelligence systems, has announced the launch of Astroware, a commercial AI safety company established by the same founding team.

Although both operate as separate entities, Astroware is structured as a customer of Trishool, purchasing red-teaming services from the subnet to support its enterprise products.

It is bringing SN23’s research and security testing capabilities into products that organizations can integrate into their AI applications.

The platform features three solutions focused on detecting unsafe AI outputs, monitoring autonomous agents, and continuously testing AI systems for vulnerabilities.

The launch also establishes a commercial relationship through which Astroware’s revenue could contribute to Trishool’s token economy.

What Is Astroware?

Astroware is an AI safety company developing tools that help businesses identify risks and introduce safeguards into their AI systems.

As organizations increasingly rely on AI for customer support, decision-making, and automated workflows, ensuring these systems behave safely is becoming more important.

AI models can generate harmful responses, while autonomous agents may execute actions that create security or operational risks. Astroware aims to address these challenges by monitoring AI behaviour and preventing unsafe actions.

Although Astroware and Trishool are separate legal entities, they share the same founding team, with Trishool providing distributed AI safety research and Astroware developing commercial applications from that work.

Inside Astroware’s AI Safety Products

Astroware currently features three products addressing different aspects of AI safety, from screening model outputs to controlling autonomous agents and identifying security vulnerabilities.

1. HaloGuard EU is a text classification model designed to detect AI outputs associated with prohibited or high-risk practices under the EU AI Act. Hosted in the European Union, it helps businesses screen generated content and support their compliance efforts.

In internal testing across all 24 official EU languages, Astroware reported 99.7% accuracy in English and 97.9% across the remaining languages.

It is approximately 100 times cheaper per message than using a general-purpose AI model with the same complete policy document.

2. Agent Auto Mode focuses on controlling autonomous AI systems by reviewing proposed actions before execution. It allows businesses to approve or block potentially unsafe operations in real time, reducing the risks associated with giving AI agents greater independence.

3. Chaos Mode provides continuous security testing through controlled adversarial attacks designed to expose weaknesses in AI systems. Successful attacks help inform improvements to existing safeguards, allowing businesses to strengthen their defenses over time.

Both Agent Auto Mode and Chaos Mode are currently available by invitation. Together, the three products aim to help organizations identify unsafe AI behaviour, control autonomous actions, and continuously improve the security of their AI applications.

How Trishool Powers Astroware

Astroware’s products draw on Trishool’s distributed AI safety research infrastructure, where participants investigate how AI systems respond to manipulation, deceptive behaviour, and attempts to bypass safeguards.

One research initiative brought together 392 participants who contributed more than 6,000 hours of work and produced 200 publicly available AI safety evaluation scenarios.

The research involved professionals from cybersecurity, medicine, finance, and law, helping identify risks that conventional testing approaches might overlook. The resulting dataset is available on Hugging Face.

Astroware is structured to purchase continuous red-teaming services from Trishool under a commercial agreement. When its products use these services, a portion of the resulting revenue will return to the subnet, with most of the initial proceeds earmarked for alpha buybacks and permanent burns or locks.

The exact revenue share is still being finalized, and buybacks will depend on enterprise adoption and revenue generation. Once they begin, the team plans to publish an onchain dashboard tracking the transactions and the revenue allocated to them.

What Comes Next for Astroware and Trishool?

Astroware represents an effort to turn Trishool’s distributed AI safety research into practical products for businesses. With HaloGuard EU as its initial offering, the company is also developing capabilities for autonomous agent monitoring and continuous vulnerability testing.

The end goal is to establish Trishool as a permissionless AI safety testing network serving multiple commercial customers, with Astroware operating as one of them.

The success of this model will depend on Astroware’s ability to attract paying customers and generate sustainable revenue for Trishool. If successful, it could demonstrate how decentralized AI research can support commercial applications while creating economic value for the subnet behind them.

Useful Resources

Follow Astroware Via:

1. Website: https://astroware.ai

2. Hugging Face: https://huggingface.co/collections/astroware/haloguard-10

3. Docs: https://astroware.ai/research/decentralized-red-teaming

4. HaloGuard EU API: https://astroware.ai/docs/haloguard-eu

5. Contact Astroware: https://astroware.ai/contact