← ObservatoryProgrammesPROG-AI

Artificial Intelligence

Claims where the contested frontier lies in whether artificial intelligence capabilities demonstrated under bounded conditions continue to hold beyond those conditions.

2026-08-01FR-AI-0002IN-007 appended — ORCA-bench (arXiv:2607.28545v1), authorised through Post-Scout flag 2026-08-01-02. Bounded CONTESTING evidence strengthens the existing low-supervision agentic boundary with operational root-cause-analysis measurements. Instance only: AS-001 remains current; Pressure State ESCALATING, Verification Stage VS-02, mechanisms, and open questions unchanged. No new assessment issued.Escalating
2026-08-01FR-AI-0007IN-008 appended — open-ended AI research shadow evaluation (arXiv:2607.27191v1), authorised through Post-Scout flag 2026-08-01-01. Bounded CONTESTING evidence: agents completed substantial research engineering but failed the tested open-ended scientific-judgment task. Instance logged before AS-002; no prior corpus content modified.Fragmenting
2026-07-17FR-AI-0007IN-007 appended — Distributed Denial of Science (arXiv:2607.10712v1). Bounded CONTESTING instance on the correctness component: under the tested adversarial retrieval setting, correctness in autonomous scientific workflows is conditional on evidential provenance. The existential claim and previously demonstrated correct discoveries are preserved; the autonomous problem-identification boundary (BN-001/OQ-001/AT-001) is unchanged. No assessment issued; pressureState FRAGMENTING, verificationStage VS-03, mechanisms, and openQuestions unchanged. The separately authorised NEW RECORD PATH component is not implemented by this mutation.Fragmenting
View Artificial Intelligence trajectoriesAI records are brought forward within the full archive.

No current programme-level diagnosis.