Warnings that advanced artificial intelligence poses a catastrophic threat to humanity are increasingly driving federal policy debates in Washington and shaping corporate lobbying strategies. In an analysis published Saturday by Al Jazeera, researcher Adji Bousso Dieng argues that this focus on existential risk is being leveraged by major technology companies to justify regulations that disadvantage open-source competitors and concentrate market power.

The debate comes at a critical juncture in the United States, where artificial intelligence has become a central issue ahead of next month’s midterm elections. Over the past week, the tension between public safeguards and industry control has surfaced in quiet Capitol Hill lobbying by AI developer Anthropic, bipartisan legislative proposals in the House of Representatives, and campaign platforms calling for outright bans on advanced AI systems.

The Debate Over Open-Source AI and Market Concentration

According to the Al Jazeera analysis, public anxiety over hypothetical doomsday scenarios has diverted regulatory attention away from immediate, documented harms. These present-day challenges include corporate layoffs driven by automation, the surging energy consumption and carbon footprint of expanding data centers, algorithmic bias, automated disinformation, and the use of AI for target identification in warfare.

Dieng argues that framing advanced AI as a potential weapon of mass destruction provides a rationale for strict regulatory frameworks that target open-source AI models, which are typically free or cheaper to deploy and can be modified by users. Prominent political figures have echoed concerns about unrestricted access, with Senator Bernie Sanders advocating for bans on superintelligence and former Secretary of State Hillary Clinton describing open-source AI as a real threat.

Training a frontier AI model from scratch now requires upwards of $100 million in compute power alone, placing sovereign development out of reach for much of the Global South. When developing nations rely on imported, proprietary models built by a handful of Western tech firms, the software frequently fails to account for local contexts—as seen in the recent failure of Kenya’s AI-driven Social Health Authority (SHA) system, which did not accurately reflect the country’s informal economy. By pushing for safety rules that restrict open-source alternatives, established tech giants risk enforcing what the analysis characterizes as digital colonization, leaving global users dependent on a few corporate providers.

Anthropic Pitches a Narrow Federal Safety Standard

While critics warn of regulatory capture, major AI developers are actively proposing their own legislative boundaries in Washington. According to Channel News Asia and Reuters, citing a Friday report by Politico, representatives for Anthropic have quietly pitched US lawmakers on a draft bill to regulate frontier AI safety and security.

An Anthropic spokesperson stated that the draft legislation reflects proposals from the company’s previously released “Advanced AI Framework” and is designed to grant the government the ability to block or deter dangerous deployments while requiring model testing, transparency, and independent evaluation. However, the specific mechanics of the draft bill place high thresholds on federal intervention.

Under the reported draft, the US government could move to block a model’s release or use only if it poses a foreseeable, imminent risk of causing more than $1 billion in economic damages, or killing or seriously injuring more than 50 people. Those risks must specifically stem from a cyberattack, the creation or release of a biological weapon, or the AI model evading its developer’s controls.

The enforcement structure relies on third-party evaluators registered with the Department of Commerce. AI labs would be permitted to pay these evaluators directly, provided the compensation is not tied to the outcome of the assessment. If an evaluator identifies an imminent enumerated risk, it must notify the Commerce Department within 72 hours. Commerce could then ask a federal judge on the District Court for the District of Columbia to halt the model. Notably, the draft language explicitly bars any other federal department, agency, or officer from suspending the development, deployment, or internal use of a model, and would override certain state-level AI laws.

Bipartisan Pushback and Electoral Politics in the US

The legislative maneuvering highlights a widening divide over how strictly the federal government should police AI labs. Under President Donald Trump, federal AI policy has relied primarily on voluntary industry pledges and executive orders aimed at sustaining an economic boom and preventing the United States from falling behind China. Yet lawmakers in both parties are increasingly pushing back against a hands-off approach, propelled by public backlash against data centers and security incidents involving autonomous AI agents.

According to an analysis by the Center for European Policy Analysis (CEPA), Democratic and Republican House members recently signaled rare consensus on the need for mandatory federal oversight, though their proposals differ significantly in scope from Anthropic’s draft:

  • The AI Kill Switch Act: Co-sponsored by Rep. Nathaniel Moran (R-TX) and Rep. Ted Lieu (D-CA), this bill would require developers of the most powerful frontier models to maintain the capability to throttle, suspend, or shut down systems if they threaten $100 million in economic damages or 10 deaths—a much lower threshold than the $1 billion and 50-death standard in the reported Anthropic pitch.
  • The FRONTIER Act: Introduced by Rep. Lori Trahan (D-MA) and Rep. Jay Obernolte (R-CA), this legislation would establish an Undersecretary for AI Security within the Commerce Department and mandate third-party evaluations for large frontier developers generating more than $5 billion in annual revenue. Like the industry-backed approach, it would preempt state laws, such as California’s independent safety audit requirements, to prevent a conflicting 50-state regulatory patchwork.
  • Liability and Minimum Standards: Trahan has also proposed the CLAIM Act to hold AI firms legally liable if their models break the law, while Rep. Don Beyer (D-VA) has floated legislation to create a Federal Reserve-like structure to mandate safety testing.

On the campaign trail, the rhetoric is even sharper. According to The Hill, Democratic Michigan Senate candidate Abdul El-Sayed recently labeled AI an existential threat, proposing an immediate pause on frontier AI development, a ban on artificial superintelligence, mandatory kill switches, and a cabinet-level AI safety agency. In the same race, Republican candidate Mike Rogers has also called for slowing down AI development to allow the government and its partners time to establish safeguards.

Key Questions Surrounding AI Safety Legislation

Would a narrow federal safety standard reduce catastrophic risks or protect incumbent developers?

Proponents of targeted federal frameworks argue that clear, high thresholds prevent regulatory overreach while establishing mandatory auditing for extreme threats like biological weapons or severe cyberattacks. Critics, however, contend that compliance costs—especially when paired with rules restricting open-source weights—primarily serve to lock out smaller startups and academic researchers, cementing the market dominance of major tech corporations.

What safeguards ensure third-party evaluator independence if AI labs pay for their own audits?

Under the draft bill pitched by Anthropic, AI developers can pay registered third-party evaluators directly as long as payment is not contingent on the assessment’s outcome. Whether this restriction is sufficient to prevent conflicts of interest remains a central policy question, as evaluators may still face financial incentives to maintain favorable relationships with the few massive labs capable of hiring them.

Who gains authority if federal legislation preempts state AI rules?

Both the bipartisan FRONTIER Act and the reported Anthropic draft would override state-level AI regulations, such as those enacted in California. While federal preemption eliminates a complex maze of 50 different state compliance regimes for software developers, it also strips individual states of the authority to enforce stricter local protections or broader liability standards if federal enforcement remains narrow.

Sources

Read More