New Senate Bill Seeks to Build a National ‘Black Box’ for AI Failures

U.S. Senators Mark Warner and Brian Schatz announced on September 24, 2026 that they will take to the Senate floor to press for passage of the Artificial Intelligence Risk Management and Security Act of 2026. The proposal responds to mounting concerns that frontier‑level AI systems could be exploited for cyberattacks or trigger serious accidents, while government oversight lags behind.

A centerpiece of the bill is a national AI incident database — essentially a “black box” for AI failures. Developers and critical‑infrastructure operators would be required to report serious security breaches, dangerous unexpected behaviors, and near misses within a specified timeframe. The goal is to give regulators, researchers, and other companies a shared evidence base to spot recurring flaws and prevent repeats of the same mistakes.

The legislation would also empower agencies such as NIST to formalize technical standards for AI risk management, testing, and security practices. For companies, that means new compliance obligations and potential exposure when incidents occur. Supporters counter that greater transparency is overdue, arguing that the most powerful models increasingly operate in safety‑critical contexts. How this bill will align with other AI proposals — and how strongly industry will resist mandatory reporting — will shape its chances of becoming the United States’ first major AI safety law.