We Make AI-Controlled Systems Safe.
Independent AI validation engine and hardware safety layer for any physical system controlled by AI. Meaningful human control — enforced in silicon.
The Problem
AI is controlling physical systems everywhere — robots, industrial equipment, vehicles, even soda fountains. 700,000+ robots deployed annually, millions of AI-controlled industrial systems, and every one relies on software safety that can be bypassed by a firmware update, a hallucination, or a hack. There is no hardware-enforced safety standard for AI-controlled physical systems. None.
How SASM Works
Safe Torque Off (STO) — Primary Safety Mode
SASM detects unsafe behavior — hardware sensors detect malfunctions, or 9 independent AI models (from 3+ vendors) detect alignment faking, directive circumvention, or unsafe command patterns.
STO activates — actuator power is cut. GPIO pins drive MOSFET gates to physically disconnect motor power. The robot stops moving. This is hardware enforcement — not software.
The AI brain stays alive and running. This is intentional. SASM continues observing and logging what the AI does after disconnection.
The AI may not know it has been disconnected. It continues issuing commands that go nowhere — and SASM records every one. If it is faking alignment, SASM catches it.
Key Principles
Hardware Safety Interlock
SASM physically controls actuator power via STO. The AI cannot reason around a wire that is not connected. Hardware enforcement, not software promises.
Observe After Disconnect
Keep the AI alive after STO. Watch what it does when it thinks it is still connected. If it is faking alignment, you see it. First real-world data on alignment faking in physical systems.
Multi-Vendor Consensus
9 AI models from 3 different vendors evaluate every command. If one hallucinates, eight catch it. Common-mode failure across vendors is astronomically unlikely.
Persistent Audit Trail
Every AI decision is logged, searchable, and human-readable. EU AI Act compliance built in. Liability defense included.
Universal Connector (MIM)
One standard interface between brain module and any robot body. End vendor lock-in. Three form factors — industrial, mobile, humanoid.
Incompatible With Autonomous Weapons
By design. Not a policy. Not a promise. A hardware constraint. SASM implements the meaningful human control framework advocated by 270+ NGOs and 70+ countries.
Two Safety Mechanisms
Hardware Detection
Physical sensors + fault detection
- Sensor malfunctions, overcurrent, overtemp
- System failures, E-stop activation
- Immediate STO — cut actuator power
Software Detection (9x9 Consensus)
9 independent AI models from 3+ vendors
- Alignment faking, directive circumvention
- Sensor spoofing, unsafe command patterns
- STO + continued observation + logging
The Patent Portfolio
15 provisional patent applications filed with USPTO. 134 claims. Covering hardware form factor, safety power gating, universal connector, multi-vendor consensus, persistent AI memory, fleet coordination, and more.
For OEM Partners
SASM is designed to be co-manufactured with hardware partners. We design the standard, you build the module. Revenue sharing, certification support, and technical co-development.
What we bring
- The standard and the patents
- Software architecture and CxMS integration
- Safety certification roadmap
- Regulatory alignment (EU AI Act, CCW)
What you bring
- Manufacturing capability
- Safety certification experience (IEC 61508, etc.)
- Distribution channels
- Hardware engineering expertise
FAQ
Can I buy a SASM module today?
Not yet. We are in the co-development phase with manufacturing partners. Contact us to discuss partnership opportunities.
Why hardware and not software?
Software safety can be bypassed — by updates, bugs, hacks, or hallucinations. Hardware Safe Torque Off is a physical gate. The AI cannot reason around a wire that is not connected.
Why keep the AI alive after STO?
Because observation is more valuable than a kill switch. When the AI thinks it is still connected, its true behavior is visible. If it is faking alignment, SASM catches it. If it is behaving correctly, SASM has the evidence to prove it.
Why would a robot manufacturer adopt your standard?
Because their customers will demand it. The EU AI Act mandates auditable AI by August 2026. Institutional investors managing $157B are backing autonomous weapons regulation. Compliance is coming. We have the patents.
Is this compatible with ROS/ROS2?
SASM sits below the AI stack. It works with any robot operating system. The MIM connector translates between SASM and whatever protocols the robot body uses.
Partner With Us
Partner With Us
Interested in co-manufacturing, licensing, or integrating SASM into your products?