NIST, IARPA Partner on Competition to Detect Trojan Attacks Against AI Systems

The National Institute of Standards and Technology is working with the Intelligence Advanced Research Projects Activity on a competition to help authorities identify and defend against threats facing artificial intelligence data and models.

In the TrojAI Leaderboard competition, participants will have multiple rounds to detect Trojan attacks against AI language models and other AI systems. After each round, NIST and IARPA will evaluate participants’ efforts and adjust the difficulty accordingly.

Information gathered from the competition will be shared with IARPA, which will identify ways to prevent future AI threats, NIST said.

The TrojAI Leaderboard program is part of NIST’s efforts to enhance cybersecurity measures.

In late August, the agency started standardizing four algorithms designed to withstand quantum-enabled cyberattacks and gave the cryptography community until Nov. 22 to submit their feedback on the standards. The algorithms NIST is working on enable quantum-secure general encryption and digital signature security.

Earlier in August, NIST sought feedback on a draft of the Cybersecurity Framework 2.0, designed to help organizations understand, address and discuss cyber risks. The final version of CSF 2.0 will be released in 2024.