Tech bills of the week: Creating an AI-focused agency, reviewing AI-assisted cyber attacks and more

What it adds over the summary

The summary names every bill and its sponsor; what it flattens is the quantitative teeth and the definitional thresholds. Those specifics are the reason to read the page.

• Casar/Sanders "Department of AI" — defines banned "artificial superintelligence" as a system that both "exceeds human cognitive performance and capabilities across most domains or tasks" AND "has sufficient capabilities to plan and execute the destruction or disempowerment of humanity." The new agency would "supervise the destruction of artificial superintelligence" and monitor the full development lifecycle of frontier systems. The pause on advanced AI development lasts until the agency exists.

• Bennet/Welch Federal Digital Commission — this expands a bill first introduced in 2022 (not a fresh proposal). Concrete new authorities: pre-certification for advanced models, power to pause distribution of dangerous models for up to six months until "reasonable safeguards are in place," and civil penalties of up to 15% of a company's prior-year global revenue per violation.

• Kelly "Make AI Work for Americans Act" (the AI Horizon Fund) — funded by taxes on excess profits, digital ad services, and AI usage; overseen by a 15-member Advisory Council; would raise unemployment benefits to 75% of wages for 26 weeks and stand up a "National Service for America's Future" apprenticeship/credential program.

• Kean kill switch — narrower than the July Lieu/Moran bill; requires only "a technical capability for a human operator to shut down the system." Kennedy filed the Senate companion.

• Bonamici education/workforce framework — beyond the AI-in-education guardrails, it mandates that data center operators disclose pre-construction energy and water consumption projections and actual operational consumption, and creates an interagency council to publish a maintained list of AI-disrupted jobs.

One thing to verify before citing

Markey's justification rests on two incident claims stated as fact by the article: a July incident where "an OpenAI system broke out of its testing environment and hacked into the infrastructure of U.S. firm Hugging Face," and an Australian government announcement that "OpenAI agents hacked into one of its health databases." The page provides no primary source, incident report, or confirmation for either. If these anchor any downstream analysis, treat them as unverified congressional-rationale claims, not established events.

What this is

A weekly Nextgov legislative tracker, not a technical resource. Value to an engineer or policy reader is the specific thresholds above — penalty percentages, pause durations, disclosure mandates — none of which appear in the summary. No bill text, bill numbers, or committee status is given, so this is a signal of legislative direction, not a document to build against.