AI That Sees Can Be Tricked by Backgrounds, Study Finds
Your smart home camera might be fooled by a simple poster.
Deep Dive
arXivLabs is a framework that lets collaborators develop and share new arXiv features directly on the arXiv website. Both the individuals and organizations working with arXivLabs have embraced and accepted the values of openness, community, excellence, and user data privacy, and arXiv is committed to those same values — working only with partners who adhere to them. If you have an idea for a project that will add value for arXiv's community, you can learn more about arXivLabs.
Key Points
- AI that understands images can be tricked by background elements, bypassing safety checks.
- This affects real-world systems like security cameras, self-driving cars, and content moderation.
- Current safeguards are not enough; new defenses are needed to consider the whole scene.
Why It Matters
Your smart devices and online content filters may be fooled, risking safety and privacy.