Hacking AI Agents: Bypassing Security in Automated Support
AI customer service agents introduce complex security vulnerabilities that allow attackers to bypass authentication and manipulate automated actions through protocol and logic exploits.
Data Poisoning
The deliberate injection of corrupted, misleading, or unusable data into AI training pipelines to degrade model quality or make data collection prohibitively costly.
Reading List
The Rise of Data Poisoning: Sabotaging the AI Slop Machines
387
Internet users are increasingly using 'data poisoning' and misinformation to sabotage AI training sets in protest of unethical web scraping.