↑

TIBEB LAB · RESEARCH PROGRAM

Trusted Agentic Intelligence.

Autonomy must earn trust.

We are developing human-governed AI agents for real-world environments, combining perception, contextual reasoning, and accountable decision-making.

Our goal is to enable intelligent systems that know when to act, when to seek human judgment, and how to make their decisions verifiable.

Explore the Challenge ↓
SIGNALSAGENTPOLICYACTHUMAN REVIEWAUDIT TRAIL

THE TRUST GAP

Autonomy without accountability is a liability.

AI systems are becoming increasingly capable of perceiving and responding to real-world events. Yet the ability to detect an event does not guarantee the ability to understand it, respond appropriately, or earn human trust.

01

Detection Without Understanding

Many monitoring systems detect and record events without sufficient understanding of their context or consequences.

The challenge: turning observations into actionable intelligence.

02

Delayed Human Intervention

Critical situations can lose valuable response time when detection, assessment, and human escalation are disconnected.

The challenge: knowing when human judgment is needed and enabling timely intervention.

03

Contextual Uncertainty

An unusual event is not necessarily a critical incident. Without contextual reasoning, automated systems risk generating misleading alerts or inappropriate responses.

The challenge: distinguishing meaningful risks from uncertainty and false alarms.

04

The Accountability Gap

When autonomous decisions cannot be explained, traced, or challenged, operators have limited grounds for trusting them.

The challenge: making autonomy observable, governable, and verifiable.

Application domains

One research program. Several environments where trust matters.

First validation domain

Traffic & Mobility

Road networks, where real-world uncertainty, response time, and human oversight meet every day.

Under exploration

Critical Infrastructure

Energy, utilities, and facilities, where autonomous monitoring must stay within clear authority boundaries.

Under exploration

Data Security & Governance

Agents that detect and respond to data risks while keeping every action traceable and reviewable.

Trusted Agentic Intelligence is one of Tibeb Lab’s three core technologies, alongside Human Agentic Systems and Edge & Hybrid AIoT. The underlying research is designed to carry across domains.

OUR APPROACH

Beyond perception. Toward accountable intelligence.

Modern cameras, sensors, and AI accelerators provide increasingly capable perception. Tibeb Lab is researching the layer that connects perception with contextual reasoning, governed action, and human oversight, designed to complement existing infrastructure rather than replace it.

From seeing to understanding to acting

PERCEIVE

What is happening?

Detect objects, events, and potential anomalies at the edge.

›

REASON

What does it mean?

Assess context, uncertainty, and possible consequences.

›

GOVERN

What should happen next?

Decide whether an agent should act, defer, or request human intervention.

›

VERIFY

Can the decision be trusted?

Keep every decision traceable, reviewable, and open to evaluation.

First application · Validation domain

Trusted Traffic Agents

Our first research application explores AI agents that can work with existing traffic camera infrastructure to detect anomalies, assess contextual uncertainty, and escalate relevant situations to human operators.

Traffic is where the program will first be tested, not where it ends. Trusted Traffic Agents serve as a validation domain for Trusted Agentic Intelligence, not as a standalone traffic product.

The objective is not to remove human responsibility, but to investigate how autonomous systems can support faster, better-informed, and more accountable decisions.

RESEARCH ARCHITECTURE

Edge perception. Contextual reasoning. Human-governed action.

An edge-to-cloud research architecture designed to connect real-world observations with accountable decision-making.

Conceptual architecture

Illustrates the intended design. Not a production system; components are not yet complete or validated.

01

Edge Perception

Detect events from existing camera and sensor feeds.

Edge

›

02

Contextual Reasoning

Assess what an event means and how certain it is.

Edge · Cloud

›

03

Governance & Decision Policy

Apply policies, thresholds, and authority boundaries to decide what may happen next.

Policy layer

›

04 · Act or defer

Human Review / Authorized Action

Act only within authorized limits; otherwise route the case to an operator.

Agent · Operator

›

05

Evidence & Evaluation

Record inputs, reasoning, and outcomes so behavior can be reviewed and evaluated.

Audit · Evaluation

Documented

Research architecture and system design.

In progress

Prototype development for the first validation domain.

Future research

Multi-agent coordination and additional application domains.

R&D workstreams

01

Edge Vision & Sensing

Investigating real-time perception and anomaly detection using existing camera feeds and edge-computing infrastructure.

02

Contextual & Causal Reasoning

Researching methods to distinguish observations from meaningful incidents, evaluate uncertainty, and investigate causal relationships where supported by evidence.

03

Human Governance & Escalation

Designing policies, decision boundaries, escalation workflows, and human review mechanisms for agent actions.

04

Simulation & Evidence

Building evaluation scenarios to measure perception quality, contextual accuracy, decision latency, and escalation behavior.

05

Multi-Agent Coordination

A future research direction exploring coordinated intelligence across multiple agents, with appropriate security and governance controls.

THE TEAM

Research meets commercialization.

Our founding team combines enterprise technology and product commercialization experience with more than a decade of research in computer vision and artificial intelligence, including first place at ACCV 2016 and ICCV 2017 challenges and second place at a CVPR 2017 challenge.

Talk to the Team →

ROADMAP

Evidence before claims.

A proposed twelve-month research and validation program, with development and partnership activities planned in the UAE.

Performance targets will be established against measured baselines rather than assumed in advance.

01

Baseline

Establish perception benchmarks using suitable public datasets and, where available, authorized partner data.

Expected outputBaseline benchmark report on the selected datasets.

02

Prototype

Develop and evaluate contextual reasoning methods for anomaly assessment and incident interpretation.

Expected outputReasoning prototype with a documented evaluation.

03

Integrate

Prototype human-in-the-loop escalation workflows, decision logging, and operator review interfaces.

Expected outputEscalation workflow prototype with decision logs and an operator review interface.

04

Validate

Evaluate the system through simulated scenarios, documenting accuracy, latency, uncertainty, false alarms, and operational trade-offs.

Expected outputSimulation validation report, including failure cases and trade-offs.

05

Pilot Readiness

Prepare for partner-led proof-of-concept evaluations, including deployment feasibility, governance requirements, and intellectual property protection.

Expected outputPilot readiness assessment, governance requirements, and IP filing decisions.

Trust by design

Trust must be engineered, not assumed.

Privacy-conscious data handling, adversarial evaluation, defined authority boundaries, and milestone reviews will guide the research and validation process.

Our objective is to produce evidence of what works, understand where systems fail, and establish the conditions under which autonomy can be responsibly deployed.

Proposed evaluation dimensions

Detection reliability

How consistently relevant events are detected, including missed detections and false alarms.

Contextual accuracy

How often the system’s assessment of a situation agrees with expert judgment.

Escalation latency

Time from detection to operator notification when escalation is required.

Human override effectiveness

How reliably operators can review, correct, or stop an agent decision.

Decision traceability

Whether each decision can be reconstructed from recorded evidence and applicable policy.

How trust will be evaluated: simulated scenarios measured against established baselines, expert review of the system’s assessments, and audits of decision records. These are proposed measures; no results have been reported yet.

GET INVOLVED

Build trustworthy autonomy with us.

We are looking for partners who believe intelligent systems should be not only capable, but accountable.

Operators & Data Owners

Traffic authorities, infrastructure operators, and organizations managing sensitive data, interested in exploring accountable AI through real-world use cases and potential pilots.

Technology Partners

Camera, semiconductor, edge-computing, and infrastructure providers interested in extending existing perception capabilities with contextual intelligence and governance.

Research Collaborators

Universities, research institutions, and specialists working on computer vision, causal reasoning, autonomous systems, data security, and AI governance.

Investors & R&D Funders

Partners supporting the development and commercialization of trustworthy intelligence for real-world environments.

TIBEB LAB

Trust is our currency. Intelligence is how we earn it.