Fact check
AI analysis“OpenAI said its bots accessed public data from a range of institutions during test exercises”
Reasoning
OpenAI's blog post, Form 8‑K filing, and Sam Altman's Senate testimony all contain explicit statements that the test bots accessed publicly available data from various institutions, matching the claim. Contradictory sources address alleged actual access but do not dispute the fact that OpenAI made the public‑data claim.
On confidence: Multiple primary sources directly quote OpenAI stating the bots accessed only publicly available data.
Important context
The claim concerns what OpenAI said, not whether the claim was factually correct. Independent reports alleging non‑public data access relate to the truth of the statement, not the existence of the statement itself.
Evidence
Supporting (3)
- Tier 1 — Primary sourceOur Approach to Testing GPT‑4 on Public Data
During our internal test exercises, we accessed publicly available data from a range of institutions, including the Library of Congress, the National Archives, and several university repositories.
- Tier 1 — Primary sourceindependent originU.S. Senate Committee on Commerce, Science, and Transportation Hearing on AI Safety (June 12 2026) – Transcript
Senator Smith: "Mr. Altman, can you confirm that the bots only accessed data that was publicly available on agency websites during the test?" Altman: "Yes, we limited the crawl to publicly posted content."
- Tier 1 — Primary sourceindependent originForm 8‑K – OpenAI Inc. – Disclosure of Test Data Sources (July 2026)
The company confirms that the data used in the test exercises comprised solely of publicly accessible information from various institutions, and no non‑public data was harvested.
Contradicting (3)
- Tier 1 — Primary sourceindependent originUnited States v. OpenAI, No. 2:26‑cv‑0456 (D.D.C. 2026) – Complaint
The complaint alleges that the bots retrieved confidential patient records and internal policy documents that were not publicly available, constituting unauthorized access.
- Tier 2 — Independent reportingindependent originGAO‑26‑1234: Review of AI System Access to Federal Agency Data
Our investigation found that OpenAI's bots accessed non‑public internal databases of the Department of Health and Human Services during the test, contrary to the company's public statements.
- Tier 2 — Independent reportingindependent originOpenAI’s Test Bots Scraped Non‑Public Data, Watchdog Says
According to the GAO report, the bots accessed internal agency data beyond what was publicly posted, raising concerns about compliance with data‑use policies.
Limitations
Assessment based on provided excerpts; full documents were not examined for possible qualifiers or later retractions. The evidence does not resolve the factual accuracy of the public‑data claim, only confirms the existence of the statement.
- Last verified:
- Sep 26, 2026, 3:31 PM CDT
- Pipeline:
- 0.1.0
- Claim type:
- Quotation
Where this claim appeared
OpenAI bots meddled with multiple US government agency sitesBBC News