California AB 316 - Ban on the "AI Decided" Defense

What Happened

California’s AB 316, which took effect on January 1, 2026, states that operators cannot use the autonomy of an AI system as a defense in liability disputes. If an AI agent causes harm, the operator cannot argue that it “did not have control over its decisions.” Colorado initially passed a similar risk-based AI Act (SB 24-205, 2024) but repealed it via SB 26-189 (May 14, 2026), replacing it with a disclosure-only ADMT regime effective January 1, 2027 (see Case 9). New York City already enforces annual bias audits for AI recruitment tools (Local Law 144).

MEG 2 Analysis - Legal Framework

Direct relevance - convergent legislative validation: AB 316 legislatively validates the central thesis of Chapter 1.1 MEG 2 through the same logic, from a complementary direction:
AB 316: eliminates autonomy defense - operator is liable, regardless of system autonomy.
MEG 2 6.1(b): autonomous decision error is a basis for liability, not exoneration; who is liable varies with the level (N1/N2/N3), but liability always exists; irreversible consequence aggravates it.
* Colorado's original risk-based provisions were repealed in 2026 (see Case 9); the AB 316 / Colorado / Texas divergence illustrates why a common technical layer of liability allocation is needed. NYC Local Law 144 is analogous to the independent audit requirements in 9.5(f).

Procedural Mechanism

Chapter 8 (Adoption as a Protocol): the 2026 retreat of comprehensive mandates (EU AILD withdrawn, Colorado risk-based Act repealed) alongside targeted measures (AB 316) illustrates Chapter 8: a voluntary, verifiable protocol is more resilient than legislative mandates dependent on consensus.
2.2 (USA - context update): Colorado AI Act updates section 2.2 of MEG 2 - from federal deregulation to active preemption and the collapse of the first state risk-based mandate.