Скачать или смотреть Data Poisoning: When Hackers Corrupt AI Before It’s Born

What if hackers didn't attack your AI—they just made sure it learned the wrong lessons from the start?". In this video, we dive into the "long game" of AI attacks: Data Poisoning. Learn how attackers inject malicious samples into training data to create hidden backdoors, causing AI to behave exactly how a hacker wants while appearing perfectly normal 99.9% of the time.
We cover:
• How poisoned data is used to sabotage autonomous vehicles, spam filters, and recommendation systems.
• The concept of "Garbage in = Compromised AI out".
• The Defense Pipeline: Why security must start before training begins through data provenance tracking, anomaly detection, and curated datasets.
Data security is AI security. If you can't verify your data origin, don't train on it.


Keywords: Data poisoning, AI security, machine learning attacks, training data, data hygiene, adversarial AI, anomaly detection, data provenance, autonomous vehicle sabotage, compromised AI.