Challenges.

Verify whether AI-generated content is authentic

Organizations cannot tell whether AI-generated content is authentic because information showing where content came from is missing or unreliable.

Prevent AI agents acting on conflicting instructions

AI agents take unintended actions when instructions conflict with external content, tool outputs, or system rules.

Block unauthorized actions by AI agents using tools

AI agents can take unauthorized actions when model decisions interact with tools, permissions, and malicious external content.

Slow false information before it outpaces fact-checkers

False information reaches audiences faster than fact-checkers can verify it because digital platforms distribute content almost instantly.

Detect AI impersonation of real people in media

AI-generated images, audio, and video can impersonate real people because they can closely resemble authentic recordings.

Protect children from unsafe advice in AI systems

Children can receive unsafe advice from AI systems because safeguards do not reliably account for age and vulnerability.

Make online privacy practices understandable before data sharing

Online services collect more personal data than users expect because privacy practices are difficult to understand before data is shared.

Locate sensitive data and who can access it

Personal data is exposed when organizations cannot identify where sensitive information is stored or who can access it.