“Do not let Sauron have the ring,” Erik Voorhees, founder of AI service Venice, wrote on Nov. 12, warning against giving politicians control over artificial intelligence. His post as @ErikVoorhees targeted the concentration of political power over machine intelligence.
His warning followed a clip of All-In co-host David Sacks predicting a Democratic push for a federal department to approve frontier AI models before release. A September Sanders-Casar bill already proposes such a department and approval powers for covered advanced AI systems. El-Sayed’s October Senate campaign platform proposes an agency that could block releases, with a different set of safeguards.
The clip published by @firesidealpha on Nov. 7 shows Sacks predicting a major Democratic push for AI legislation in the next Congress. He describes a federal AI department reviewing the most advanced models, giving developers its concerns and withholding release until those concerns are addressed.
Sacks goes further, alleging that Democrats want AI-powered surveillance and censorship. The proposals establish release-blocking powers, not the political motives he alleges.
The remarks came from the Nov. 7 All-In episode, announced by @theallinpod. In the full episode, Sacks also predicts a bargain in which government gains more power while closed-model companies gain regulatory protection. His concern therefore reaches beyond censorship to the possibility that compliance rules could protect established developers from competition.
Voorhees argues that open-source transparency and decentralization provide better safety, and that licensing undermines equality. His objection is to anyone holding that gatekeeping power.
A department and approval requirement are already proposed
Sen. Bernie Sanders and Rep. Greg Casar announced the Ban Artificial Superintelligence Act on Sept. 23, before Sacks’s November remarks. The introduced Senate version, S.5493, would establish a Department of Artificial Intelligence, pause advanced AI development pending a regulatory framework and prohibit artificial superintelligence.
Its proposed approval mechanism is explicit. Section 9(d) would prevent a person from deploying, releasing, importing or transferring an advanced AI system before receiving the department’s approval. The secretary would inspect for characteristics posing a danger to the public, including superintelligence and precursor characteristics, and withhold approval if dangerous characteristics are present.
The bill would also require a charter to develop or distribute advanced AI. Maintaining that charter would entail giving regulators access to systems, processes, staff and physical infrastructure. Separate provisions would require pauses and shutdown measures for specified dangerous systems.
This would put a federal decision between a covered developer and release of its model. It resembles the institutional mechanism Sacks describes much more closely than a voluntary testing arrangement does. It remains proposed legislation, however, rather than evidence that the next Congress has agreed to create such a system.
The bill also contains constraints on the regulator. It proposes Senate confirmation of the secretary, independent scientific advice and employee financial-interest restrictions. It provides Federal Circuit review of findings underlying charter revocations. That specific remedy is not a general appeal of every model-release decision.
Abdul El-Sayed’s Oct. 13 “AI Under Control” platform calls for an independent, Cabinet-level agency empowered to inspect laboratories, block releases, order pauses and command shutdowns. The campaign document explicitly builds on the Sanders-Casar legislation.
It proposes freezing training and release of the most powerful systems until safety rules and enforcement exist, with independent scientists certifying safety before the pause lifts. It also calls for independent pre-release audits and mandatory breach reporting to government within 24 hours, accompanied by public summaries.
The safeguards are part of the proposal, too. A scientific board chosen outside the agency would manage its governance. Exemptions would be public, industry fees would fund the regulator and officials would be barred from later cashing in on the companies they regulate. The platform also supports whistleblower protections and preserving stronger state laws.
Crucially for Sacks’s surveillance allegation, the document reiterates opposition to warrantless surveillance. Its stated prohibitions also include autonomous weapons firing and replacing human oversight in life-or-death decisions. The proposed release powers do not establish Sacks’s allegation about the purpose they would serve.
Those protections do not eliminate the concentration of authority Voorhees opposes. An agency able to stop a release would retain that power even with outside scientific oversight. The safeguards seek to constrain its use, while leaving the regulator with a veto over covered releases.
The platform does not set out a detailed release-approval or appeal procedure. Its outside-selected scientific board also differs from S.5493’s advisory board, whose members the secretary would select. The two proposals overlap in coercive powers while making different choices about who advises or governs the regulator.
Who would constrain the regulator?
Introduced legislation: Senate confirmation, scientific advice, financial restrictions and review of charter-revocation findings.
Campaign proposal: an outside-selected scientific board, public exemptions and limits on officials profiting from regulated companies.
Release-blocking powers are already proposed. The bill and campaign platform put different safeguards around that authority.
Both are proposals, not an enacted common system. S.5493 covers defined advanced AI. The Federal Circuit route shown concerns findings behind charter revocations, not general release-approval appeals. The campaign platform does not detail an approval or appeal procedure.
What changes after the election
Sacks’s reference to the next Congress comes against an election shift. CryptoSlate, citing Associated Press, reported that Democrats had won the House and unified control of Congress in its election coverage, updated Nov. 8.
Existing federal policy also needs to be distinguished from the proposed approval regime. Trump’s June 2 executive order directs a voluntary framework for government access to frontier models. Section 3(c) expressly says that section does not authorize mandatory government licensing, preclearance or permitting for new models. That limitation concerns the framework in that order, rather than every other government power over AI.
For Voorhees, the issue is longstanding and tied to his business. His July 2024 introduction to Venice identifies him as founder and CEO, describes an application using leading open-source models and argues against corporate or government monopoly over intelligence. CryptoSlate also covered the connection between model-access restrictions and decentralized AI in June.
The next legislative test is whether these proposals gain committee action and leadership backing, and what any new text says about release criteria, decision-making authority and review of adverse decisions.