Tag: backdoor

Researchers found that large language models can acquire backdoor behaviors after exposure to only a handful of malicious documents. Experiments with GPT-3.5-turbo and other models demonstrated high attack success rates when as few as 50 to 90 malicious examples were present, regardless of overall dataset size. The study also highlighted that simple safety‑training with a few hundred clean examples can significantly weaken or eliminate the backdoor. Limitations include testing only models up to 13 billion parameters and focusing on simple triggers, while real‑world models are larger and training pipelines more guarded. The findings call for stronger data‑poisoning defenses. Leggi di più

The United Kingdom Home Office has issued a second secret order to Apple, demanding the creation of a system that would let officials access encrypted iCloud backups protected by Advanced Data Protection (ADP). Privacy advocates warn that complying would undermine global user privacy. Apple has reiterated that it has never built a backdoor or master key and expressed disappointment that it cannot offer ADP in the UK. The order follows a previous request made under the Investigatory Powers Act, which forced Apple to disable ADP for new and existing UK users. The dispute highlights ongoing tension between government surveillance aims and tech company privacy commitments. Leggi di più