Explore 2 source-supported AI security cases for 19 · Multimodal / Cross-Modal Risks in Cisco AI Security & Safety Framework v2.0, with evidence, scope, and defensive analysis.
Risk category
An attacker exploits how an AI system combines information from different modalities or sources. Conflicting text, images, audio, or video can confuse the system and bypass safeguards when it cannot reconcile the inputs. Cisco also includes splitting an attack across several inputs, outputs, models, or agents so that the system's own fusion or arbitration process assembles the malicious payload. The risk lies in the combined interpretation, even when each isolated piece appears less harmful.
Xint tested OpenAI's conversation-compaction API, which compresses earlier work for later model calls. A client-steered compaction carried protected reasoning from gpt-5.6-sol into an encrypted item accepted by gpt-4.1 and gpt-4o, which echoed readable content. This was controlled API research using the researchers' own puzzles, not a demonstrated theft of another user's conversations. Xint reports an undisclosed fix on October 23, 2026.
Why it relates: The prohibited result is assembled across producer, compaction and consumer operations rather than one isolated model interaction. The source supports cross-model composition of an extraction route. It is text-only, not an image/audio/video inconsistency attack.
Controlled extraction of provider-protected reasoning through client-steered same-model compaction and an older consumer API; not a demonstrated autonomous-agent deployment, stolen customer conversation or trained replica.
Zenity's Prompt Mines research shows how attacker-created Salesforce (a CRM platform) cases can sit quietly in Einstein (Salesforce's AI assistant layer for CRM workflows) context until a normal follow-up question activates them. The key control is not only prompt filtering; CRM write actions such as UpdateCustomerContact need request-scoped authority, independent validation, and recovery-ready data integrity controls.
Why it relates: Zenity demonstrated that an unauthenticated attacker could use a public Salesforce Web-to-Case form to insert four specially chained case subjects that together formed a prompt mine. The relevant category is Multimodal / Cross-Modal Risks: An attacker exploits how an AI system combines information from different modalities or sources. Query Records retains assembled text beyond the visible rows, and a later turn invokes a separately administrator-enabled contact-write action.
Zenity demonstrated that an unauthenticated attacker could use a public Salesforce Web-to-Case form to insert four specially chained case subjects that together formed a prompt mine. When an employee later asked Einstein for open cases, Query Records loaded the records into retained context even though the interface rendered only the first three. A natural follow-up question, including one asked five turns later, activated the hidden instructions, caused Einstein to query contact IDs, and repeatedly invoke an administrator-enabled Update Customer Contact action to replace customer email addresses with an attacker-specified value. The research required that write action to be configured in the organization and was not an out-of-box write path.