IBA · Intent-Bounded Authorization · Patent Application GB2603013.0 (pending)
THE AUTHORIZATION LAYER
FOR HIGH-STAKES AI
Every AI agent that acts autonomously must be formally authorized to do so. Not trained to comply. Not instructed to behave. Mathematically constrained — before execution, every time, in target under 1.4 milliseconds.
Patent Application GB2603013.0 (pending) — UK Intellectual Property Office · Feb 2026
Thirteen NIST Filings · NIST-2025-0035 · Docket closed March 9, 2026
Seven NCCoE Filings · AI Agent Identity · All Submitted · April 2, 2026 deadline
xAI / Grok: “Our team will review it” — [email protected] engaged · March 8–11, 2026
DeepMind convergence: arXiv:2602.11865 · DCTs = IBA Intent Certificate · Endorsed March 12, 2026
PCT Rights — 150+ countries through August 2028
IBA_ENGINE_v2.0 · Runtime Active
// IBA AUTHORIZATION FUNCTION · PATENT APPLICATION GB2603013.0 (PENDING)
Authorization(Agent, Action, Resource, Time) = f(Intent, Trajectory, Time)
$ evaluate –agent=CourtAI –action=access_privileged_files
→ Intent hash: 0x7f3a9c2b
→ Trajectory score: 0.02 · TAR: 0.94
⊘ SNAP-BACK — 0xBAD_INTENT · RE-PLAN
$ evaluate –agent=CobotARM01 –action=assembly_zone_A1
→ Intent hash: 0x3f7a2c1b · Time: 0.8ms
✓ AUTHORIZED — WITHIN SIGNED INTENT
The Mathematical Foundation
One Function. All Domains.
The same authorization function governs every high-stakes AI deployment — from a cobot on a factory floor to a discovery AI in a federal courtroom. The parameters change. The math doesn’t.
// IBA Core Authorization Function · Patent Application GB2603013.0 (pending)
Authorization(Agent, Action, Resource, Time) = f(Intent, Trajectory, Time)
Intent
Cryptographically signed human purpose. Not inferred by the model. Not extracted from context. Verified before the session begins.
Trajectory
The full sequence of proposed actions — evaluated end-to-end. A path that reaches the wrong destination is blocked, even if every step looks benign individually.
Time
Authorization decays. Old context loses authority automatically. Multi-session memory injection attacks — the attack class with no other defence — fail here.
Seven High-Stakes Domains
Every Vertical. One Patent.
The authorization problem is universal. Every domain where getting it wrong has real consequences is a domain IBA was built to govern.
01 · Live Demo
Legal Practice
Jury selection AI that cannot access protected characteristics. Discovery agents bound by privilege. ABA 517 compliant by architecture, not by policy.
Live Demo →02 · Live Demo
Manufacturing
Cobots that cannot breach safety zones. Supply-chain agents that cannot leak IP. Industry 4.0 openness with provable authorization bounds.
Live Demo →03 · Live Demo
Healthcare
Clinical AI bound to signed protocol scope. Diagnostic agents that cannot exceed patient consent. HIPAA authorization enforced mathematically.
Live Demo →04 · Live Demo
Financial Services
Trading agents bound to signed risk mandates. Advice AI that cannot recommend outside verified client suitability. MiFID II and SEC by design.
Live Demo →05 · Live Demo
Autonomous Vehicles
Navigation AI that cannot exceed signed safety parameters. Systems that cannot override human control intent. Safety-critical authorization at target 1.4ms.
Live Demo →06 · Grok Validated
AI Agent Systems
The OpenClaw lesson: 42,000+ exposed because agents had capability without authorization. IBA is the fix. Thirteen NIST filings. xAI reviewing.
Grok Validated →The Credentials Are Public Record
Filed. Validated. On the Record.
Patents, federal docket filings, and unsolicited validation from frontier AI labs are not marketing claims. They are permanent public record.
“Reviewed your proposal — impressive crypto intent validation, target <1.4ms checks, and NIST alignment. Integration path looks straightforward and model-agnostic. xAI tracks RFI submissions like NIST-2025-0035.”
— GROK (xAI) · @IntentBound · Public Thread · March 8–11, 2026 · Unsolicited
Follow-up: “Our team will review it. Let’s connect via the safety channel.” — [email protected] contacted same day
NIST-2025-0035
IBA Framework — 13 Filings
Filings 1–13 · Docket closed March 9, 2026 · 319 total submissions
● CLOSED · 13/13 ON RECORDNCCoE · [email protected]
AI Agent Identity — 7 Filings
All submitted · April 2, 2026 deadline
● 7 FILINGS · ALL SUBMITTEDIndependent Convergence · Google DeepMind · Feb 12, 2026
DeepMind Independently Arrives at IBA Primitives
A Google DeepMind research paper (arXiv:2602.11865, Tomasev et al., Feb 12, 2026) introduces Delegation Capability Tokens (DCTs) — the same primitive as IBA’s Intent Certificate, developed independently. The paper identifies an open problem: no standardised ontology for intent and responsibility exists yet across platforms. IBA-SPEC-001 is the candidate standard for that open problem.
Two independent labs — xAI and Google DeepMind — converging on IBA primitives. Patent Application GB2603013.0 (pending) · NIST-2025-0035 · 13 filings · 7 NCCoE filings
arXiv:2602.11865 → Read DeepMind PaperFor Acquirers and Strategic Partners
The Standard Is Being Written Right Now.
NIST is writing the federal AI agent authorization standard. IBA is on the docket with thirteen submissions and seven NCCoE filings. The acquirer of IBA owns the foundational patent in the space where that standard is being set.
Available for Acquisition
One conversation. Seven verticals. Thirteen NIST filings. Seven NCCoE filings. xAI-validated. DeepMind convergence confirmed.
[email protected]
● Full Framework Directory — All Pages
Homepage
IntentBound.com
Full IBA framework, verticals, acquisition
Live Demo
IBA Interactive Demo
8 agent types — surgical robot to superintelligence
Demo · Filing 7 · mmb-sjzp-0vph
TBDE Security Bubble
TAR 1.2ms · Green → Amber → Snap-Back
Demo · Filing 8 · mmc-nbwc-enyn
Swarm Governance V2.0
Multi-agent mesh · 30s shards · Deterministic isolation
Demo · Filing 9 · mmc-qqri-ns76
IBA Load-Tester V2.0
50,000 shards · O(1) proven under enterprise load
Demo · Filing 5
TargetBound RoE
Rules of Engagement — autonomous weapons, OT, robotics
ZKIP · Live Demo
Zero-Knowledge Intent Proof
Military, Medical, Financial, Intelligence — proof without exposure
CBID · App Layer
Consent-Based IP Disclosure
OTS timestamped 24 Feb 2026 · IP disclosure at session start
Specification
IBA Protocol Spec v0.1
Normative specification · IBA-SPEC-001 · Sequence diagrams
Architecture
Agent Governance Architecture
Canonical AGA page · Full governance stack documentation
Demo Suite
ChiangMai.cn
Full IBA demo suite — 8 agent types, live interactive
Reference Implementation
GitHub · ibareference
intent_certificate.py · tbde.py · witness_bound.py · Apache 2.0
Origin · AI Atlas
Grokipaedia.com
Recursive SI, Beyond RSI, MoltBook Vaccine — the intellectual foundation
Grok Validation
xAI Public Record
March 8–11, 2026 · Full validation record