A Handful of Bad Data Can 'Poison' Even the Largest AI Models, Researchers Warn

A handful of bad data can 'poison' even the largest AI models, researchers warn

A new study undermines the conventional wisdom that greater scale means greater safety.

Anthropic Study Reveals Alarming AI Poisoning Attack Risk - American Faith

Researchers collaborating with Anthropic AI have demonstrated a troubling vulnerability in large language models: a “poisoning attack” using just 250 malicious documents can make these systems produce nonsensical output when triggered. The study was conducted alongside institutions like the Alan Turing Institute and the UK AI Security Institute. Poisoning attacks work by covertly inserting corrupt or misleading examples into a model’s training d…

Anthropic Study: AI Models Are Highly Vulnerable to 'Poisoning' Attacks

A recent study by Anthropic AI, in collaboration with several academic institutions, has uncovered a startling vulnerability in AI language models, showing that it takes a mere 250 malicious documents to completely disrupt their output. Purposefully feeding malicious data into AI models is ominously referred to as a "poisoning attack." The post Anthropic Study: AI Models Are Highly Vulnerable to ‘Poisoning’ Attacks appeared first on Breitbart.

How many malicious docs does it take to poison an LLM? Far fewer than you might think, Anthropic warns - WorldNL Magazine

Just 250 corrupted files can make advanced AI models collapse instantly, Anthropic warnsTiny amounts of poisoned data can destabilize even billion-parameter AI systemsA simple trigger phrase can force large models to produce random nonsenseLarge language models (LLMs) have become central to the development of modern AI tools, powering everything from chatbots to data analysis systems.But Anthropic has warned it would take just 250 malicious docu…