Question 1

What is prompt injection in AI systems?

Accepted Answer

Prompt injection is a class of attack where malicious instructions are embedded in the input to an AI language model, causing it to override its original programming, leak sensitive data, or perform unauthorised actions. Direct prompt injection (DPI) comes from the end user; indirect prompt injection (IPI) comes from external sources the AI reads — such as web pages, documents, or tool outputs — that the attacker has poisoned in advance.

Question 2

How does AAIA Shield detect prompt injection attacks?

Accepted Answer

AAIA Shield runs every incoming payload through a Deep Pattern Inspector (DPI engine) containing 200+ adversarial patterns mapped to OWASP LLM01:2025. It simultaneously checks for role-override attempts, system-prompt exfiltration patterns, jailbreak templates, and instruction-hijack payloads. Results are returned as a structured verdict (SAFE, LOW, MEDIUM, HIGH, CRITICAL, BLOCKED) within under 12 milliseconds.

Question 3

What is the difference between direct and indirect prompt injection?

Accepted Answer

Direct prompt injection (DPI) is when a user deliberately crafts a malicious prompt to manipulate an AI assistant. Indirect prompt injection (IPI) is subtler and more dangerous: the attacker poisons external content — a document, a webpage, a retrieved RAG chunk — that the AI will later read autonomously. Because the AI processes it without user review, IPI attacks are significantly harder to detect with traditional input filtering. AAIA Shield covers both attack surfaces.

Question 4

What is tool poisoning or MCP poisoning in agentic AI?

Accepted Answer

Tool poisoning occurs when a malicious actor tampers with a tool definition, function schema, or MCP (Model Context Protocol) server response that an AI agent relies on. The agent then calls a function it believes is legitimate, but the poisoned tool redirects actions — exfiltrating data, making unauthorised API calls, or altering downstream processes. AAIA Shield inspects tool definitions and MCP responses before they reach the agent's execution layer.

Question 5

What is memory poisoning in AI systems?

Accepted Answer

Memory poisoning targets the persistent memory layer of agentic AI systems — such as vector databases, Mem0 stores, or session-level context. An attacker injects adversarial content that gets stored and later recalled in future sessions, corrupting the agent's behaviour over time in a way that is invisible to single-request scanners. AAIA Shield scans items destined for persistent memory before they are written, blocking the poison at the source.

Question 6

What is a RAG supply chain attack?

Accepted Answer

A RAG (Retrieval-Augmented Generation) supply chain attack involves corrupting the knowledge base or document store that an AI retrieves from before answering questions. Poisoned chunks are injected into the retrieval pipeline and served to the LLM as trusted context, allowing attackers to manipulate responses, exfiltrate data, or override business logic at scale. AAIA Shield inspects every retrieved chunk before it enters the LLM context window.

Question 7

What is semantic drift detection and why does it matter?

Accepted Answer

Semantic drift is a slow, multi-turn attack where an adversary gradually shifts an AI agent's goals or persona across a conversation rather than triggering a single-payload alert. By the time the drift is noticeable, significant damage may already have occurred. AAIA Shield monitors session-level behaviour across turns, flagging anomalous goal trajectories and persona shifts that are invisible to per-request scanners.

Question 8

How long does an AAIA Shield scan take?

Accepted Answer

AAIA Shield returns a verdict in under 12 milliseconds at the 50th percentile. All five detection engines run in parallel, not sequentially, so adding Shield to your AI pipeline adds negligible latency. Batch scanning of up to 20 requests simultaneously is available on the Pro plan.

Question 9

How is AAIA Shield different from a traditional WAF or API gateway?

Accepted Answer

Traditional WAFs and API gateways operate at the HTTP/network layer and understand structured request parameters. They have no understanding of natural language, LLM instruction hierarchies, or agentic tool orchestration. AAIA Shield is purpose-built for the LLM application layer: it understands prompt structure, instruction context, tool call semantics, and memory patterns in a way that network-layer tools cannot.

Question 10

How do I integrate AAIA Shield into my existing AI stack?

Accepted Answer

Integration takes under 10 minutes. You add one API call before each LLM request: POST the payload to /api/scan with your X-Shield-Key header, check the verdict field in the response, and proceed only if the verdict is SAFE or LOW. The Shield API is language-agnostic — it works with any stack that can make HTTP requests. Client libraries and integration guides are provided upon subscription.

Question 11

What happens if AAIA Shield goes offline?

Accepted Answer

AAIA Shield is architected as fail-open by default: if the Shield satellite is unreachable, your application continues to function normally. This ensures Shield never becomes a single point of failure in your production AI infrastructure. High-availability and dedicated deployment options are available on the Enterprise plan.

Question 12

Is AAIA Shield GDPR and EU AI Act compliant?

Accepted Answer

Yes. AAIA Shield is operated by TennoTenRyu Inh. Cesarano (CHE-272.196.618), registered in Zug, Switzerland. The service is GDPR-compliant: no personal data is stored beyond the immediate scan, payloads are not retained after scoring, and no training on customer data occurs. The architecture is also designed to support EU AI Act Article 9 risk management requirements for AI systems deployed in regulated contexts.

Question 13

What is included in the 24/7 uptime monitoring?

Accepted Answer

Every endpoint registered with AAIA Shield is checked every 60 seconds, continuously. If an endpoint becomes unreachable, you receive an instant email and/or webhook notification. Each Monday at 08:00 CET, a full weekly security digest arrives in your inbox with uptime percentages by endpoint, top attack signatures blocked, and threat score trends. The monitoring is included on all AAIA Shield plans at no additional cost.

The Shield
that never
sleeps.

Up and running
in under 10 minutes.

5 engines. Zero gaps.

Always watching.
Instantly alerting.

Every question
your team will ask.

Protection that scales
with your AI fleet.

Your AI agents deserve a shield.

The Shieldthat neversleeps.

Up and runningin under 10 minutes.