AI training data poisoning
PulseAugur coverage of AI training data poisoning — every cluster mentioning AI training data poisoning across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
New theory analyzes AI data poisoning in continual learning
A new theoretical framework has been developed to analyze data poisoning attacks and defenses in continual learning (CL). Researchers framed the interaction between adversaries and defenders as an online zero-sum game, …
-
New latent class attack and detection method detailed in arXiv paper
Researchers have developed a new data poisoning technique called a latent class attack, which introduces a novel, unknown class of data and mislabels it as a known class. This attack could be used to bypass AI-based sec…
-
Bilevel optimization framework detailed for Neural Architecture Search
This paper provides a structured overview of Neural Architecture Search (NAS) by framing it as a bilevel optimization problem. It categorizes existing NAS methods into sampling-based and bilevel theory-based approaches.…
-
LLM Security Threats Evolve Beyond Traditional Defenses
The rapid integration of Large Language Models (LLMs) into various industries has introduced a new set of security challenges that traditional methods struggle to address. Key vulnerabilities include prompt injection, w…
-
New attack targets robot learning via world model vulnerabilities
Researchers have identified a new vulnerability in robot learning pipelines that exploit world models. By injecting malicious prompts or compromising transition dynamics into seemingly safe datasets, attackers can creat…
-
New Cordyceps attack enables covert control of LLMs via data poisoning
Researchers have developed a new data poisoning technique called Cordyceps, which enables covert control attacks on large language models (LLMs). Unlike previous methods that rely on fixed trigger phrases, Cordyceps tea…
-
Data poisoning disrupts BigTech surveillance and AI training
Data poisoning is a method to disrupt the data used by large technology companies for surveillance and AI training. This technique involves subtly altering or corrupting data inputs to mislead AI models. By introducing …
-
AI blueprint analysis poses hidden security risks
A security analysis highlights the risks associated with AI systems that interpret engineering blueprints, such as those developed at Skoltech. These systems, which use multimodal models to read and analyze architectura…
-
AI Security and Observability Guides for 2026 Released
The provided articles offer a comprehensive guide to AI application observability and security testing for the year 2026. They detail methods for identifying and mitigating unique AI security threats such as prompt inje…
-
Video promotes data poisoning to disrupt AI systems
A YouTube video advocates for data poisoning as a method to disrupt AI systems. The content suggests this approach as a form of resistance against the proliferation and capabilities of artificial intelligence.
-
AI data poisoning concerns grow with large language models
The concept of "data poisoning" in AI models is being discussed, particularly in relation to large language models trained on vast datasets like Wikipedia. This issue highlights concerns about the integrity and reliabil…
-
Data poisoning can trick surveillance algorithms, exposing personalization fragility
Data poisoning is a technique that can be used to manipulate algorithms, particularly those involved in mass surveillance and personalization. By strategically introducing corrupted data, individuals can potentially tri…