The National Institute of Standards and Technology (NIST) released an updated guide detailing various adversarial machine learning attacks and mitigation strategies. The report emphasizes AI's vulnerability and differentiates threats between predictive AI systems, which include evasion, data poisoning, and privacy attacks, and generative AI models, which face supply chain and prompting attacks. Notable changes from the initial document include the addition of generative AI models' learning stages and comprehensive classifications of attacks, addressing the growing complexity of AI-related cybersecurity threats.
AI is useful but vulnerable to adversarial attacks. All models are vulnerable in all stages of their development, deployment, and use. At this stage with the existing technology paradigms, the number and power of attacks are greater than the available mitigation techniques.
Some of the substantial changes in the final guidelines from the initial version released in January 2024 include... an index on the classes of attacks on different AI systems.
For predictive AI systems, the NIST guidelines review evasion attacks, data poisoning attacks and privacy attacks, all of which change the underlying data powering AI models.
For generative AI models, the three listed attacks are: supply chain, direct prompting and indirect prompt injection, each with methods to corrupt a model's output.
Collection
[
|
...
]