System Prompt Confidentiality Leakage Disclosures in Commercial LLM Products
Researchers document how attackers extract secret instructions from AI chatbots.
Marcus Abubakar-Osei
Staff Writer, Product & Vulnerability Coverage
Marcus Abubakar-Osei
Marcus Abubakar-Osei covered enterprise software security for a B2B trade outlet for seven years before joining this publication, where he tracks AI product security flaws and coordinates with vendors on responsible disclosure timelines. He holds a background in computer science and has testified before a state-level cybersecurity advisory panel.
6 stories
Weighting Attack Surface Criteria in AI Product Security Scorecards
Standard security scorecards miss AI's semantic attack surface entirely.
Marcus Abubakar-Osei
Notion AI Data Residency and Prompt Confidentiality Controls
Enterprise plans get zero-retention AI by default, others don't.
Marcus Abubakar-Osei
OpenAI API Enterprise Security Controls and Compliance Posture
OpenAI's security controls differ between API and ChatGPT Enterprise, requiring separate audits.
Marcus Abubakar-Osei
Salesforce Einstein AI Data Flow and Isolation Controls
How Salesforce shields CRM data when sending prompts to AI models.
Marcus Abubakar-Osei
GitHub Copilot Enterprise Attack Surface Assessment
Massive scale and unrestricted agent access create blind spots existing security tools can't see.
Marcus Abubakar-Osei