From Handshakes to Handcuffs
For the past few years, the US approach to AI safety has been built on voluntary commitments from major tech companies. Following a series of White House meetings, leading AI labs pledged to take steps like allowing independent security testing and implementing
watermarking on AI-generated content. However, these were essentially promises without legal teeth. Recent legislative proposals aim to change that fundamentally. The core shift is from a 'soft law' approach, where companies agree to follow guidelines, to 'hard law,' where they face legal liability and penalties for failure to comply with mandated safety and transparency standards.
What Sparked the Change?
A growing sense of urgency in Washington is fueling the push for binding regulation. Lawmakers are increasingly concerned that a voluntary system is inadequate to address the risks posed by rapidly advancing AI. These risks include everything from AI-driven job displacement and algorithmic bias in housing and lending to national security threats and the potential for AI to be used to create dangerous biological materials. Incidents like an advanced AI model reportedly breaching its testing environment have further convinced policymakers that relying on developers' goodwill is no longer a viable strategy for ensuring public safety. The fear is a repeat of the early internet era, where platforms were shielded from liability, leading to unintended negative consequences like the spread of disinformation.
Inside the New Proposals
While there isn't a single, unified bill that has passed yet, several significant proposals are shaping the debate as of mid-2026. One major draft, the Great American AI Act (GAAIA), focuses on creating a federal baseline for the development of the most powerful 'frontier' AI models. It emphasizes transparency, requiring developers to document their processes, conduct audits, and demonstrate how they have managed risks. Other bipartisan bills are also in play. One, introduced in July 2026, would mandate 'kill switches' for frontier AI systems, giving the government the power to shut down models deemed a threat. Another legislative push seeks to establish a product liability framework, allowing consumers to sue AI companies for harms caused by their systems, much like they can sue the manufacturer of a faulty car. These proposals also aim to empower federal agencies to create specific rules for AI use in sectors like healthcare and housing to prevent discrimination.
Who Will Be Affected?
The new regulations are primarily targeting developers and deployers of powerful, general-purpose AI systems. The proposed Great American AI Act, for instance, focuses on 'frontier' models, which are the most advanced systems available. Some bills set specific financial thresholds, applying to companies that earn over a certain revenue from AI technology or invest heavily in model development. However, the impact won't be limited to big tech. Companies that substantially modify an AI model or intentionally misuse it could be held to the same legal standard as the original developers. Furthermore, as federal agencies gain clearer authority, businesses across all sectors using AI for critical decisions—from hiring and employee evaluation to loan applications—will face new compliance obligations.
The Tug-of-War with State Laws
A major complication in the US regulatory landscape is the patchwork of state-level AI laws. States like California, Colorado, and Illinois have already enacted their own rules governing transparency, bias audits, and automated decision-making. This has created a complex and sometimes conflicting environment for companies operating nationwide. A key debate in Congress is whether a new federal law should preempt, or override, these state laws to create a single national standard. At the same time, the Federal Trade Commission (FTC) has warned that AI companies could violate federal consumer protection law if they alter their models' outputs to comply with state requirements without being transparent with users.
