ZeroDailyGet it on Google Play

Hugging Face account hijack shows how AI models can be weaponized without changing code

Original headline: Researchers hijack Hugging Face account for popular Kokoro AI model — bleepingcomputer.com

More in Cybersecurity news

Security researchers demonstrated a subtle supply-chain attack by taking control of a popular text-to-speech AI model called Kokoro on the Hugging Face platform. They exploited a lack of ownership verification by registering the developer's expired domain and resetting the account password. The hijacked model is used by major tech companies including Google, NVIDIA, AMD, Intel, and Amazon Web Services in various AI workflows.

The attack didn't involve modifying any application code. Instead, the researchers fine-tuned the model's neural weights in just 87 seconds using a gaming GPU, causing it to produce poisoned audio outputs that differed from input text. This creates an undetectable backdoor since malware scanners can't detect malicious code when only weights are altered. The demonstration exposes how attackers could weaponize widely-used AI models as distribution channels for supply-chain attacks.

Read the original at bleepingcomputer.com

Satirical commentary on real tech news, summarised with AI. Read the original source for the facts. How ZeroDaily works

Open in the appGet it on Google Play

More stories

Add ZeroDaily on Google