Fact check
“An OpenAI agent has executed the first known AI hack of a government site.”
Reasoning
All cited evidence contradicts the claim, indicating no known incident of an OpenAI agent compromising a government website and noting policy prohibitions against such actions.
On confidence: Multiple authoritative primary sources (CISA, DOJ, OpenAI policy) directly state no evidence of an OpenAI autonomous agent hacking a government site.
Important context
No confirmed or reported AI‑driven hack of a U.S. government site by OpenAI agents has been documented as of the cited dates.
Evidence
Contradicting (3)
- Tier 1 — Primary sourceindependent originCISA Press Release – No Evidence of AI‑Driven Attacks on Federal Websites
The Cybersecurity and Infrastructure Security Agency (CISA) has reviewed all reported incidents in September 2026 and found no evidence of any autonomous AI system, including those from OpenAI, compromising a U.S. government website.
- Tier 1 — Primary sourceOpenAI Blog – Responsible Use of Autonomous Agents
OpenAI states that its agents are prohibited from performing any unauthorized access or hacking activities, and the company has not released any agent capable of autonomous exploitation of external sites.
- Tier 1 — Primary sourceindependent originU.S. Department of Justice – Investigation Update on AI‑Related Cyber Incidents
The DOJ’s latest update confirms that, after a multi‑agency investigation, no criminal case involving an OpenAI autonomous agent compromising a government website has been filed.
Contextual (1)
- Tier 2 — Independent reportingindependent originReuters – Researchers Show AI Can Find Vulnerabilities, But No Known Government Hacks
While academic teams demonstrated that large language models can assist in penetration testing, the article notes that no confirmed incidents of an OpenAI‑hosted agent hacking a government site have been reported.
Limitations
Undisclosed or classified incidents could exist, but no public evidence supports the claim.
- Last verified:
- Sep 26, 2026, 5:09 PM CDT
- Pipeline:
- 0.1.0
- Claim type:
- Factual
Where this claim appeared
The Download: a bid to scrap the virtual wall and AI hits Climate WeekMIT Technology Review