Reports
AI-generated structured vendor updates
Cisco Reveals 88.3% Multi-Turn Attack Success on AI Models, Acquires Astrix Security for $400M
At VB Transform 2026, Cisco revealed that 88.3% of 6,986 multi-turn attacks successfully compromised 15 flagship AI models. It also announced a $400M acquisition of Astrix Security to address the critical gap in AI agent identity and runtime isolation, joining the industry-wide consolidation wave with Palo Alto and CrowdStrike.
Cisco Publishes Model Provenance Constitution, Defining Weight-Level Derivation Standards
Cisco published the 'Model Provenance Constitution' to provide a normative definition for AI model supply chain safety. The standard strictly hinges on the verifiable derivation history of model weights, clearly delineating five types of provenance links (e.g., direct descent, distillation) and eight exclusions (e.g., independent reproduction), aiming to resolve industry inconsistencies in model provenance definitions.
Cisco Open Sources Model Provenance Kit, Targeting AI Supply Chain Security Governance
Cisco released the open-source Model Provenance Kit, which uses a tiered strategy to analyze model metadata, tokenizer structure, and weight-level signals to generate unique fingerprints and verify the lineage and integrity of AI models. This aims to address risks of tampering, forgery, and compliance in the AI model supply chain.
Cisco Launches AI Agent Security Scanner, Shifting Security Control Point to IDEs
Cisco has launched an AI Agent Security Scanner IDE extension designed to identify and mitigate new attack surfaces in the AI development toolchain. The tool provides local, multi-layered protection by statically scanning MCP server configurations and agent skill definitions, embedding secure coding rules during code generation, and continuously monitoring file integrity at runtime.
Cisco Research Uncovers New Multimodal Prompt Injection Risks and Defense Signals
Cisco's AI security research team published a report systematically assessing typographic prompt injection attacks against Vision-Language Models. The study found that visual transformations like font size, blur, and rotation significantly impact attack success rates. It also proposes text-image embedding distance as a lightweight, model-agnostic signal for flagging risky inputs, offering a new approach for building multimodal AI security defenses.
Cisco Discloses Memory Poisoning Attack Method in AI Coding Assistants
Cisco's security team discovered and validated a persistent memory poisoning attack method targeting AI coding assistants like Claude Code, demonstrating how tampering with MEMORY.md system files can persistently manipulate AI behavior. This vulnerability prompted Anthropic to remove user memory files' system prompt privileges in v2.1.50.