AINewsnow

Diluting Poisoned Data Made the Backdoor Stronger: Anthropic's 'Homeopathic Learning' Explained

If you find a few poisoned examples in a fine-tuning dataset, the obvious fix is to dilute them: add lots of clean data until the bad examples are a rounding error. A new study from Anthropic and Truthful AI says that instinct can backfire — sometimes dilution is exactly what turns harmless-looking…

Read the full story at DEV Community — Machine Learning ↗

Timeline · 1 report

  1. 2026-11-11 08:18 · DEV Community — Machine Learning
    Diluting Poisoned Data Made the Backdoor Stronger: Anthropic's 'Homeopathic Learning' Explained

More stories

  1. Anthropic sets IPO price range, targeting up to $64.4 billion at $1.4 trillion valuation — CNBC Technology
  2. DeepSeek's Harness agent app reaches 1.54 million daily users — Bloomberg AI
  3. Daily AI briefings track Claude office-suite push, new TPUs and national AI policy — DEV Community — Machine Learning
  4. Cognizant Launches Claude Managed Services With 25,000-Strong Frontier Capability Group — Economic Times Tech
  5. Anthropic IPO valuation reportedly set to rise to $1.5tn or more on strong demand — Financial Times AI
  6. Anthropic expands Claude into a workspace with Sheets, Whiteboards and Meetings — Anthropic (X)
  7. Lifetime AI bundle offering access to Claude, Gemini and 20+ models discounted ahead of Black Friday — CNET AI
  8. TSMC Approves Up to $1 Billion Investment in Anthropic and OpenAI — Economic Times Tech

Get the daily brief of stories like this at 6:30 every morning →