First impression: This lit review boiled down to availability, integrity, abuse, and privacy. Poisoning and evasion. The motivation seems to be to formalize attacks so that another AI can respond, e.g. GANs. AVs are sketchy. Deepfake was left out. They already have competing supply chains and sanctions or economic interventions. It does not really get into the yet-to-be-released versions like for robotics. What makes ML different? Or Generative AI? This identifies flaws in existing systems and infrastructure, not what the next would be. Other versions might be evolutionary or deterministic. Look at personas and proxies. Some of this would come out of the ways they already beat tech. Capitalism continues to disrupt itself, but those become new standards. The same transcript might be interpreted by copilots as consumers, clinics, or culprits. "I didn't know anything about crypto when I first got involved. I barely knew what a blockchain was." -- SBF.
Lessons From Red Teaming 100 Generative AI Products
View PDF
Influence and cyber operations an update october 2024
Microsoft Secure Future Initiative september 2024 progress report
Simple probes can catch sleeper agents \ Anthropic
LLM Agents can Autonomously Exploit One-day Vulnerabilities
ComPromptMized
LLM Agents can Autonomously Hack Websites
Download PDF
NIST Artificial Intelligence (AI) 100-2 E2023, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations
Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations
Csa living off the land
National cybersecurity strategy 2023