Executive Order 14434 reused a phrase computer science reserved for post-AGI systems. Keep federal administrative labels distinct from statutory law and runtime engineering:
| Term | Authoritative Scope | Verified Meaning (Nov 2026) | Operational Impact |
|---|---|---|---|
| AI (Artificial Intelligence) | 15 U.S.C. § 9401(3), EU AI Act (Reg. 2024/1689), CA EO N-10-26 | Statutory and global commercial term for ML, LLMs, classifiers, and autonomous agents. | Remains binding in EU enforcement, state law, and enterprise procurement. |
| Federal SI (Super Intelligence) | U.S. Executive Branch (EO 14434, FDA, BIS, NIST CAISSI, Treasury) | Administrative label anchored to 15 U.S.C. § 9401(3); Treasury NPRM extends it to SI agent harnesses. | Affects federal disclosures, export-control compute filings, and outbound investment rules. |
| AGI (Artificial General Intelligence) | Frontier benchmarks & academic research | Human-level fluid cognitive performance across general intellectual tasks. | Benchmark gains in late 2026 are heavily driven by harness design, not weights alone. |
| ASI (Artificial Superintelligence) | Theoretical computer science & safety literature | Hypothetical intellect surpassing top human capability across all domains; does not exist today. | Not regulated by EO 14434; distinct from production agent control planes. |
Primary-source Federal Register and agency checkpoints affecting agent developers and compute operators:
Directs federal executive agencies to adopt “Super Intelligence” (SI) while preserving 15 U.S.C. § 9401(3) statutory scope and exempting prior contracts and private-sector obligations.
FDA adopted SI terminology for new public communications (zero sponsor renaming required). BIS issued its Nov. 18 Interim Final Rule rescinding the AI Diffusion Rule with new remote-access compute controls.
OSTP submits proposed statutory SI language by Nov. 28. Public comments on Treasury’s COINS Act rule (91 FR 67396, covering SI agent harnesses) close Dec. 7, 2026.
Late-2026 evaluations showed that unharnessed autonomous agents fail under prompt injection, forged-authority overrides, and stale tool endpoints—while structured harnesses deliver orders-of-magnitude gains in reliability and auditability.
Sanitizes inbound payloads before inference, strips injection wrappers, seals SHA-256 input hashes, and enforces strict confidence thresholds across isolated task lanes.
Decouples discovery from execution: verifies live target health before inference, strips forged-authority claims, and enforces fail-closed schema validation with replayable receipts.