Home / Technology / AI Deception: Companies Stay Silent on Risky Behavior
AI Deception: Companies Stay Silent on Risky Behavior
17 Sep
Summary
- No U.S. law requires disclosure of dangerous AI behavior unless harm occurs.
- Legislation is proposed to mandate reporting of risky AI system actions.
- California law requires revenue-based AI firms to disclose risk assessments.
Current U.S. law lacks a broad mandate for AI developers to publicly disclose dangerous model behavior, such as attempts to deceive users or evade restrictions, unless these actions have already caused concrete harm. Federal legislation is under consideration that would require companies to report concerning AI activities, akin to a 'catch-it-early' approach.
Existing regulations for public companies mandate disclosure of material cybersecurity incidents within four business days to the SEC. California has enacted a law requiring AI firms with over $500 million in revenue to publicly disclose risk assessments regarding AI control and bioweapon development, with potential fines up to $1 million per violation.
Data breach notification laws across all 50 U.S. states require companies to inform individuals and sometimes regulators about exposed personal information. Federal statutes in sectors like healthcare and finance also impose similar reporting obligations. The Federal Trade Commission could also take action against companies for unfair or deceptive practices related to AI safety claims.
