Static API Endpoint
FC-2025-004: AutoGPT Indirect Prompt Injection Shell Escape CISA KEV Listing
FC-2025-004
VERIFIED TRUE
2025-06-01T00:00:00Z90 Days (Deadline: 2025-08-30)Agentic AI & LLMs
2025-08-12T16:00:00Z1. Probabilistic Hypothesis & Calibration Metrics
Section titled โ1. Probabilistic Hypothesis & Calibration Metricsโยซ CVE-2025-26319 will be added to the CISA KEV catalog within 90 days of initial disclosure due to observed threat-actor usage against enterprise autonomous agents. ยป
1.0 (TRUE)
0.0484
(P - O)ยฒ (Random baseline: 0.25)+81%
2. Ground Truth Resolution Oracle
Section titled โ2. Ground Truth Resolution OracleโUnder Hermes Epistemic Standard (Principles P5 & P10), a prediction cannot be self-judged. Its outcome is adjudicated solely by verifiable third-party ground-truth telemetry:
โ๏ธ ORACLE TYPE: cisa_kev_inclusion
Verification Source: CISA KEV Catalog
Deterministic Adjudication Criterion:
Inclusion in CISA KEV with verified CVE-2025-26319 identifier before 2025-08-30T00:00:00Z.
3. Epistemic Rationale & Evidence Chain of Custody
Section titled โ3. Epistemic Rationale & Evidence Chain of CustodyโPre-Event Rationale Formulation
Section titled โPre-Event Rationale FormulationโAutonomous agents with unrestricted workspace write access and internet-retrieval plugins provide an ideal target for indirect prompt injection leading to host shell escalation.
Supporting Inferences & Forensic References
Section titled โSupporting Inferences & Forensic Referencesโ๐ Inference Pointer: INF-AUTOGPT-EXEC-01
๐ Evidence Reference: https://www.cisa.gov/known-exploited-vulnerabilities-catalog
๐ Evidence Reference: /cve/2025/cve-2025-26319/
4. Mathematical Brier Score Verification
Section titled โ4. Mathematical Brier Score VerificationโIn accordance with Section 24 and Section 78 specifications, predictive calibration is scored under quadratic penalty:
Brier Score = (P - O)ยฒ- Forecast Probability (P):
0.78 - Realized Outcome (O):
1 - Calculated Score:
0.0484 - Skill Score Improvement:
+81%vs uninformative prior (BS_random = 0.2500).
5. Machine-Readable API Endpoint
Section titled โ5. Machine-Readable API EndpointโInspect the raw JSON telemetry artifact for this prediction:
# Direct HTTP GET fetch of the audited forecast objectcurl -s https://hermes-codex.vercel.app/api/forecast/FC-2025-004.json | jq .Global Forecast Catalog