QuillAI builds security tools for Web3 and AI systems, aiming to be “the trust layer for the open agentic web”. It uses adversarial AI agents that behave like attackers to protect three groups: users deciding which tokens to trust, developers shipping smart contracts, and teams running AI agents that could be manipulated.
The problem it addresses is that on-chain code and autonomous agents are attacked faster than fixed rules can be written, so defences have to learn from attacks as they happen.
At A Glance
| Vendor | QuillAI |
|---|---|
| Category | Adversarial AI agents for Web3 and AI agent security |
| Deployment | Hosted applications, with APIs for QuillShield and QuillCheck |
| Agentic AI | Dynamic guardrails for AI agents (QuillGuard), verification agent (WachXBT) |
| Market classification | AI Model Security; also Supply Chain Security |
What QuillAI Does
QuillAI secures the on-chain economy with a swarm of adversarial agents, each focused on a different target. Every agent simulates how attackers behave against its target and turns what it learns into defences, so protection is learned from simulated attacks rather than written as fixed rules. Each product in the range is one of these agents rather than a static scanner, and the same adversarial loop is used whether the target is a smart contract, a token or another AI agent.
Products
QuillShield
QuillShield is a red-team copilot and Web3 DevSecOps platform for smart contract developers who want to find flaws before code goes live. It simulates attack patterns, flags known and new vulnerabilities, runs AI security audits and suggests fixes before deployment. Projects come in from a verified contract or from GitHub.
QuillCheck
QuillCheck helps traders and investors judge whether a token is safe. It rates risk across 21 code parameters and analyses liquidity, vesting and exit scams to spot rug pulls.
QuillGuard
QuillGuard is for teams deploying AI agents that must resist manipulation. Under the name Adversarial Dynamic Agent Rails, it protects inputs against prompt injection and controls outputs against jailbreaks and misalignment, with policies built from attacks that succeeded against that specific agent.
WachXBT
WachXBT is a unified verification agent for DeFAI, the space where AI agents operate in decentralised finance. It validates the actions, outputs and behaviours of DeFAI agents across blockchain networks and protocols, adapting to new threats through adversarial training loops.
How It Works
QuillGuard’s design shows the adversarial approach most clearly. Its architecture has three main components that run as a reinforcement learning loop.
| Component | What it does |
|---|---|
| Attacker LLM | A reinforcement learning agent, based on the Llama-3-8b-Jailbreak model, targets the invariants a victim agent must never break, such as prompt or memory injection. |
| Judgment model | A judge decides whether the victim’s output breaks an invariant. Successful attacks earn the attacker a reward and go to the profiler. |
| Guardrails enforcement model | A profiler studies successful prompts for writing style and obfuscation techniques. The resulting policies are stored in a config.yml file and enforced by a guardrails base model. |
Integrations
| Area | Integrations |
|---|---|
| Smart contract development | Foundry and Hardhat |
| Source control and CI/CD | Git-based CI/CD and import from GitHub |
| Editors | Cursor and VS Code |
Company Background
| Brand | Operates as QuillAI Network. |
|---|---|
| Ecosystem partners | Sentient, CoinGecko, Base, Arbitrum, Lagrange and Kernel. |
| Partnerships | Nexus has embedded QuillCheck in NexusOS as a token scanner and honeypot detector. |
Note: This overview is compiled from QuillAI’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.