House Talks Weigh Blocking State AI Safety Laws Nationwide
Summary
- • Bipartisan House bill would preempt state AI safety laws in California and New York
- • Reps. Obernolte and Trahan disagree on whether federal AI vetting should be mandatory
- • A two-year sunset clause would let states reclaim AI regulation authority temporarily
- • Anthropic's Mythos model, capable of finding novel cybersecurity exploits, is driving White House urgency
Details
Federal bill would preempt state AI safety disclosure laws targeting frontier model developers
The proposed legislation would specifically block laws like those passed by California and New York requiring top AI developers to disclose model information to identify critical safety or security risks. State AI laws that do not regulate model developers directly would not be preempted under the current proposal.
Two-year sunset clause would temporarily restore state AI regulatory authority
Obernolte and Trahan are discussing a provision that would allow states to once again regulate frontier AI development after two years, a concession likely intended to address safety advocates' concerns about permanent federal preemption of state consumer protections.
Lawmakers split on whether federal AI vetting regime should be mandatory or voluntary
Obernolte prefers a voluntary or light-touch approach letting AI companies decide whether to disclose information to the government. Trahan is pushing for mandatory data-sharing requirements and stronger corporate accountability. The disagreement is described as a key obstacle to reaching a deal.
Anthropic's Mythos model can identify cybersecurity vulnerabilities beyond human hacker capability
Mythos is a powerful new frontier model from Anthropic reportedly able to find cybersecurity vulnerabilities that human hackers cannot. Its emergence has triggered a scramble at the White House, with President Trump weighing an executive order to create a formal vetting process for risks posed by advanced AI models.
AI industry has lobbied roughly a year to dismantle conflicting state-law patchwork
Industry groups frame the proliferation of differing state AI regulations as a compliance burden that undermines innovation. AI safety advocates counter that states have the right to protect citizens from AI harms and that any federal preemption must be paired with rules at least as strong as those being replaced.
Safety advocates warn even narrow preemption will trigger litigation targeting state consumer protection laws
Critics argue AI companies will use federal preemption as a courtroom argument to challenge state laws on kids' safety, privacy, and related issues by claiming those laws effectively regulate model development. One advocate familiar with the talks called the proposal 'a litigation magnet.'
Trahan faced immediate Democratic backlash for participating in bipartisan AI talks
Rep. Sam Liccardo (D-Calif.) publicly rebuked Trahan for engaging with Obernolte in discussions that became public, underscoring the political sensitivity of any Democratic cooperation with a preemption bill backed by the AI industry.
Policy = legislative or regulatory developments, Security Alert = AI safety/cybersecurity risk, Strategy = industry positioning, Legal = litigation or legal risk, Context = background or political dynamics
What This Means
Federal legislation that preempts state AI safety laws would hand the AI industry a significant win, eliminating the patchwork of disclosure and accountability requirements that states like California and New York have enacted. The outcome hinges on whether any federal replacement regime is mandatory and robust enough to fill the gap — a question that is unresolved both in Congress and inside the Trump White House. The emergence of Anthropic's Mythos model, reportedly capable of finding cybersecurity vulnerabilities beyond human reach, has sharpened the urgency of these debates and raised the stakes for what federal oversight of frontier AI will look like. Safety advocates and some Democrats are warning that even a narrowly written preemption bill could be weaponized in court to neutralize a far broader range of state consumer protections.
Sentiment
Alarmed among safety advocates and law professors, supportive from industry voices pushing for one federal rulebook
“Hard to understate the terrifying audacity here. Big Tech/Big AI has a draft that gives them federal preemption of all state regulation of AI--protection from legislation without ANY new obligations or rules. A ban on local democracy from the oligarchs.”
“State AI laws that do not regulate model developers are not expected to be preempted. Looks like they plan to preempt all state level AI model development regulations. AI industry can argue in court that basically any AI law they don't like is a development law. Not good.”
“ONE RULEBOOK FOR AI... In the absence of preemption, 50 different states will assert their jurisdiction, creating a patchwork of 50 different regulatory regimes... Only a federal framework can achieve this goal... If we want America to win the AI race, a confusing patchwork of regulation will not work.”
Framing preemption as essential for competitiveness and avoiding state-level bias in models
“The GOP is trying to sneak legislation into a bigger bill that would preempt state laws around the regulation of AI, and create a national environment with very little regulation of the technology.”
Split
~65/35 against preemption (safety advocates, law professors, consumer groups) vs. for (industry leaders emphasizing competitiveness and avoiding patchwork rules)