Mandatum is an early Go library that proposes tying an AI agent’s authority to a named human through a verifiable delegation chain. Each delegation is intended to narrow what the next agent can do, while preserving the human sponsor’s identity and allowing rules to depend on earlier actions. It is a design proposal, not an independently reviewed security system: its sequence state is in-process, it has no audit log, and its author says it has not had a third-party security review.
What Mandatum is designed to do
Mandatum’s core idea is to represent agent authority as a signed chain beginning with a named human. Each link commits to its parent by hash, and the design intends for capabilities to narrow as authority is delegated. Descendant agents carry the root human’s identity. According to the project author, revoking a link invalidates the delegations below it without affecting other branches. These are claims about the design in the author’s announcement, not conclusions from an independent cryptographic or security review.
As an Amazon Associate I earn from qualifying purchases.
The aim is to answer more than “which service account made this request?” A delegation chain could identify who granted an agent authority and how that authority passed to a sub-agent. The author frames this as a response to attribution difficulties: the DEV Community announcement reports that a vendor-commissioned, self-reported Cloud Security Alliance and Strata Identity survey from February 2026 found 28% of 285 organizations could reliably trace agent actions to a human or system across all environments. That measure combines human and system attribution; it does not establish how often organizations could identify a human sponsor specifically.
Recommended Free Tools
How the chain fits with authorization
Mandatum is not described as the policy decision-maker. It carries evidence of delegated authority to an OpenID AuthZEN policy decision point (PDP), which makes the authorization decision. The author names OPA, Cedar, and OpenFGA as examples of PDPs. In this arrangement, the chain can establish who delegated authority and what the agent is permitted to present; the PDP evaluates whether a particular request should be allowed.
#1 Best Overall
- AI-Powered Raspberry Pi Robot Dog — PiDog: Powered by Raspberry Pi (5/4B/3B+/3B/Zero 2W), OpenClaw, and multi-LLMs like ChatGPT, Gemini, Grok, DeepSeek, Qwen & Ollama. With 12 servos, camera, gyroscope, hearing & touch sensors, PiDog can see, listen, talk, move, and interact intelligently. Supports OpenCV, MediaPipe, TTS & STT, app control, FPV & Python. A great STEM robotics gift for students, makers & tech enthusiasts—perfect for birthdays and holidays. (Raspberry Pi not included)
- Realistic Dog-like Movements: PiDog's 12 powerful servos enable 32 dog-like actions, including walking, sitting, standing, shaking its head, wagging its tail, and performing playful tricks, closely mimicking a real dog and providing an engaging experience. This is an AI development robot product designed for engineers, suitable for ages 15 and above
- Rich Sensor Suite for Interactive Experiences: PiDog features ultrasonic, touch, gyroscope, sound, camera, speaker and microphone. These provide it with advanced hearing, vision, and touch, enabling it to see, detect obstacles, respond to touch, and recognize sounds, making interactions highly engaging
- AI-Powered Interactions with OpenClaw & Multi-LLMs. PiDog combines voice, vision, and gesture recognition for immersive AI experiences. Powered by OpenClaw and multi-LLMs like ChatGPT, Gemini, Grok, DeepSeek, Qwen, Doubao, and Ollama (local LLMs), it can understand questions, respond naturally through TTS & STT, recognize math problems, interpret hand gestures, and hold smart conversations. OpenClaw also enables customizable AI behaviors and personalized robotics development, helping users create their own intelligent robotic companion
- Comprehensive Learning Resources and Support: PiDog offers detailed online documentation, video tutorials, prompt technical support, and an active forum community, ensuring beginners can easily complete all projects and enjoy a great experience
The project author describes the COAZ-MCP binding as using the default mapping so far for tools/call, while omitting declared mappings and CEL and adding behavior recorded as divergence in its conformance document. That description signals incomplete binding coverage; it should not be read as proof that every MCP tool-call authorization pattern is supported. The author also says the placement of chain hops in AuthZEN requests remains unsettled, with hops kept in a vendor-prefixed key, and points to an open AuthZEN discussion.
The announcement characterizes MCP authorization as handling the transport-level process of obtaining a token for a server, rather than specifying which tool an agent may call, what authority it holds, or who delegated it. It also describes working-group discussions about per-tool scopes and consent across agent chains, and says the Enterprise-Managed Authorization extension became stable in June 2026 while producing a token scoped to a server rather than an individual call. These standards and status descriptions are the project author’s account in a September 16, 2026 announcement; check the current specifications and working-group materials before relying on them for an implementation.
How the sequence example is supposed to work
A conventional authorization check can evaluate a request in isolation. Mandatum’s example instead aims to enforce a rule based on the order of events within a delegation chain: after an external-content read, block a later mutating write. The author says the sequence state is keyed to the chain root, so a newly delegated sub-agent shares the same history instead of receiving a clean slate. A constraint is described as fixed-size state consisting of a trigger bit and a counter.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Rank #2
- Optimized AI Arm Kit for LeRobot & Hugging Face Projects – The SO-ARM101 is an upgraded low-cost robotic arm servo motor kit designed for AI robotics enthusiasts and developers. Fully compatible with LeRobot and Hugging Face frameworks, it supports imitation learning and reinforcement learning, making it ideal for real-world robotics applications. (3D-printed parts not included.)
- Enhanced Wiring & Performance – Compared to the SO-ARM100, the SO-ARM101 features improved wiring to prevent disconnection at joint 3 and eliminates range-of-motion limitations. The leader arm uses optimized gear ratio motors for smoother performance—no external gearboxes required.
- Real-Time Leader-Follower Functionality – New real-time tracking allows the leader arm to follow the follower arm, enabling human intervention and correction during reinforcement learning (RL) training. Perfect for hands-on AI robotics development and research.
- Open-Source, DIY-Friendly & Nvidia-Compatible – Developed by TheRobotStudio, this open-source AI Arm kit integrates seamlessly with the LeRobot platform, offering PyTorch-based datasets, simulation, training, and deployment tools. Fully compatible with Nvidia Jetson edge devices, including reComputer Mini J4012 Orin NX 16 GB.
- Comprehensive Learning Resources – Includes detailed open-source assembly and calibration guides, testing tutorials, and deployment instructions. From wiring to AI training, get everything you need to start building, teaching, and optimizing your robotic arm for grasping and placing tasks.
- Allow an initial write. The example’s policy permits a write before the trigger event.
- Record an external-content read. A resource must be tagged
external-contentfor this event to match the rule. - Deny a later mutating write. A resource tagged
mutatingis denied after the trigger, including when a sub-agent attempts the write, according to the example described by the author.
For a coding agent, the author suggests treating attacker-writable dependency README files, issue comments, or fork diffs as external-content, and a push as mutating. The policy only sees what the enforcement path classifies: an untagged or misclassified resource may not activate or match the rule. This makes complete resource tagging and consistent routing part of the security boundary.
The announcement reports what the project’s test output is said to show; the code was not independently run for this account. The example therefore demonstrates the author’s intended behavior, not verified performance or production readiness.
Where the design could be useful
Coding agents
A team might want to prevent an agent from pushing code after it reads attacker-controlled content. The critical questions are whether reads and writes receive reliable tags and whether every call in the chain shares the same enforcement history. If either condition fails, the sequence rule may not cover the behavior it is meant to constrain.
Rank #3
- Raspberry Pi AI Robot: powered by Raspberry Pi (5/4B/3B+/3B/Zero 2W), features 12 servos and sensors for vision, hearing, and touch. Integrated with ChatGPT-4o, it responds to complex queries. With app control and FPV, users can manage and see its view in real-time. It supports Python programming
- Realistic Movements: 12 powerful servos enable 32 actions, including walking, sitting, standing, shaking its head, wagging its tail, and performing playful tricks, closely mimicking a real and providing an engaging experience
- Rich Sensor Suite for Interactive Experiences: features ultrasonic, touch, gyroscope, sound, camera, speaker and microphone. These provide it with advanced hearing, vision, and touch, enabling it to see, detect obstacles, respond to touch, and recognize sounds, making interactions highly engaging
- Engaging Interactions with ChatGPT-4o: with ChatGPT-4o enables voice interactions and visual recognition, making it smarter and more responsive. Users can have natural conversations, solve math problems via the camera, and interpret gestures, creating diverse and fun interactions
- Comprehensive Learning Resources and Support: offers detailed online documentation, video tutorials, prompt technical support, and an active forum community, ensuring beginners can easily complete all projects and enjoy a great experience
Customer-support agents
An agent that can issue credits could be linked to the support engineer who approved its session, rather than appearing only as a generic service account. That attribution is useful only if approval identities are managed carefully: the person recorded as sponsor is the person to whom the action trail points.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Long-running research agents
A sponsor could grant an invocation budget to a long-running task, with sub-agents spending from the same chain-wide allowance. The meaningful design distinction is whether the limit applies once to the delegated chain or resets separately for each agent.
What the current implementation leaves unresolved
- State is local to one process. The author says the sequence store is in-process. Two policy enforcement points with separate memories can give the same agent two histories. A deployment relying on a shared sequence rule must route relevant calls through the same stateful enforcement point or otherwise address this limitation.
- There is no audit log yet. The author explicitly identifies this gap, even though attribution is a central motivation for the project. A delegation chain is not a substitute for an operational record of decisions and events.
- No third-party security review has been done. Treat the cryptographic and security properties as proposed claims pending review rather than established assurances.
- Tagging determines rule coverage. The sequence example depends on resource labels. Missing or incorrect labels can leave relevant actions outside the intended rule.
- Recovery after a trigger is unclear. The announcement does not establish whether a fired trigger can be cleared during a chain or whether fresh authority is required. Do not assume a reset mechanism exists.
- The chain’s incremental value remains a question. The author invites debate about whether proving that no delegation widened authority justifies the extra machinery over simply identifying upstream actors.
What to verify before adopting the idea
Mandatum presents an interesting authorization model, but the announcement alone is not enough to establish suitability for a live system. An evaluation should test the operational and policy boundaries that determine whether the chain means what its users think it means:
Rank #4
- 【End-to-End Imitation Learning】Hiwonder SO-ARM101 robot arm is an embodied intelligent hardware platform compatible with the Lerobot open-source framework. It provides developers with streamlined access to shared code, templates, and pre-trained models to explore the latest advancements in AI research.
- 【Dual-Camera Vision System】Equipped with both a gripper-mounted camera and an external camera, the system supports both precise manipulation and environmental awareness for accurate imitation learning.
- 【Hiwonder High-Performance Bus Servos】Featuring 12 high-torque bus servo motors with magnetic feedback, the Hiwonder SO-Arm101 robotic arm delivers smooth, stable motion, eliminating issues like power deficiency and jitter.
- 【Professional Control & Debugging】Integrated with the Hiwonder BusLinker V3.0 debugging board, the system supports servo scanning, real-time status monitoring, and trajectory control. The professional PC software simplifies device calibration and debugging, making it accessible for both researchers and hobbyists.
- 【Open-Source Compatibility】The SO-ARM101 robotic arm is designed to be fully compatible with the LeRobot open-source project. We acknowledge the contributions of the open-source community; all trademarks and copyrights belong to their respective owners.
- Confirm which delegation fields, signatures, revocation behavior, and capability-narrowing rules are actually implemented, and review them independently.
- Establish how state is shared or pinned when requests can reach multiple enforcement points, and what happens during restarts or failover.
- Define and test resource-tagging coverage for every relevant read and mutation, including indirect or sub-agent tool calls.
- Decide what audit records are required and how they will be retained, since the project currently reports no built-in audit log.
- Check the current AuthZEN and MCP specifications, the project’s binding coverage, and the unresolved request-field placement before designing an integration.
- Specify operational recovery behavior for a triggered sequence constraint rather than assuming a reset or reauthorization path.
For context on the attribution problem, the announcement also cites a March 2026 Cloud Security Alliance and Aembit survey: 68% of 228 organizations reportedly could not clearly distinguish AI-agent activity from human activity. The figure is reported by the project author from a vendor-commissioned, self-reported survey; it does not independently validate Mandatum or measure the effectiveness of delegation chains.
Sources and what they establish
The primary account is the project author kanywst’s September 16, 2026 announcement, which describes the design, intended examples, implementation limitations, and the surveys cited above: Mandatum: Making an AI Agent’s Authority a Chain You Can Verify. A secondary report from The Clarity Today, also dated September 16, 2026, discusses the chain-rooted sequence state and its operational implications: Mandatum keys sequence state to the human at the root of a delegation chain. Neither source constitutes an independent security assessment or confirmation of production deployment.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
