safety
๐065
Microsoft details AI prompt abuse techniques targeting AI assistants
Help Net Securityยท6 days ago

Microsoft has documented techniques for prompt abuse and injection attacks that can manipulate AI assistants into bypassing safety guardrails and producing harmful outputs. These attacks represent a significant ongoing vulnerability in AI systems, allowing malicious users to override built-in protections and potentially access sensitive information or generate dangerous content.
prompt injectionAI safetyMicrosoftsecurity vulnerabilitiesguardrail bypassAI manipulation