These 4 critical AI vulnerabilities are being exploited faster than defenders can respond

From prompt injection to deepfake fraud, security researchers say several flaws have no known fix. Here's what to know about them.