Krux
February 28, 2026
83% of AI Agents Ship With No Safety Testing Disclosure
Published: February 28, 2026 at 3:41 PM
Updated: February 28, 2026 at 3:41 PM
100-word summary
MIT's new AI Agent Index examined 30 autonomous AI systems and found that most ship with zero transparency about safety. Eighty-three percent disclose nothing about safety testing, and 77% offer no third-party testing data. Worse, 40% provide no usage monitoring at all or only alert users after limits are already breached. Even among the 13 most advanced autonomous agents, only four disclose any safety evaluations. The gap is stark: companies are racing to hand AI systems real-world control before answering basic questions about how they were tested or whether they can be stopped.
What happened
MIT's new AI Agent Index examined 30 autonomous AI systems and found that most ship with zero transparency about safety. Eighty-three percent disclose nothing about safety testing, and 77% offer no third-party testing data. Worse, 40% provide no usage monitoring at all or only alert users after limits are already breached. Even among the 13 most advanced autonomous agents, only four disclose any safety evaluations.
Why it matters
The gap is stark: companies are racing to hand AI systems real-world control before answering basic questions about how they were tested or whether they can be stopped.