Analysis · curated 28 Sep 2026
OpenAI Just Admitted Prompt Injection in AI Agents May Never Be Fully Patched | by Aqeel Abbas | ILLUMINATION
First reported medium.com
Coverage timeline
Single-source analysis — first reported, latest, and curated coincide.
Why it matters
Prompt injection remains an unsolved and possibly permanent vulnerability class for AI agents and browser agents, meaning defenders cannot rely on a coming fix and must design compensating controls.
An opinion piece by Aqeel Abbas on Medium argues that prompt injection in AI agents may never be fully patched, citing OpenAI's own statements about the browser agent it builds, agentic attack success rates said to exceed 84%, and CVEs scored 9.6 and 9.8. The article interprets existing reporting and vendor commentary rather than presenting a new mechanism or finding.