# Live news

Canonical: https://ournews.ai/live/b3161642-c1b4-4968-bef4-1c369d4235f4
Published: 2026-11-11T01:08:05.271455+00:00
Updated: 2026-11-11T01:08:05.271455+00:00

Adding clean data can strengthen hidden AI backdoors, Anthropic-backed research finds in small-model experiments. The effect depends on contrasting data and a trigger; the study did not test production training pipelines.

Source: Anthropic / Truthful AI research — https://alignment.anthropic.com/2026/homeopathic-learning/
Source date: 2026-11-10