SECAI Core · phase 11 of 15
Output Integrity Attack
A security attack where the attacker changes or interferes with the final output of a machine learning model
The Explain card
- Plain English
- An output integrity attack changes or interferes with the final output of a machine learning model after it has been produced, so consumers receive a result the model never actually gave.
- Example
- Malware on an inference server rewrites fraud-detection verdicts in transit, flipping "suspicious" to "approved" for selected transactions, while the model itself remains untouched.
- Why it matters
- A perfect model is useless if the pipe after it is tampered with. Defenders sign or authenticate model outputs, protect the inference path, and monitor for verdicts that disagree with expected patterns.
- Hook
- The oracle spoke the truth; the messenger rewrote it.
Where it sits in the deck
Phase 11: Threat Landscape: Attack Vectors and Adversarial Techniques
With defences named, learn what they must defend against — the full catalogue of attack techniques targeting AI systems, their inputs, outputs, training pipelines, and supply chains.
