Version 4

Demis Hassabis (Google DeepMind) proposed a FINRA-style U.S. Frontier AI Standards Body for pre-release testing of frontier AI models, drawing initial backing from Microsoft, OpenAI, and Chamath Palihapitiya [6]. OpenAI has since put forward CAISI — a mandatory federal evaluation framework through NIST — which has advanced legislatively with a $20M funding plan [9], but OpenAI and the White House have formally diverged: the Trump administration reportedly favors a voluntary approach closer to Hassabis's model [11]. This creates a three-way split among the proposal's original supporters. The most direct challenge to Hassabis's credibility as a safety proponent is Google DeepMind's own Pentagon contract, which violated DeepMind's founding safety conditions; an internal petition of 250+ employees failed to stop it [14].

The split between OpenAI's mandatory federal design and the White House's preference for a voluntary approach moves the debate from policy papers to active legislative and executive negotiations — one with real stakes, given that the CAISI bill's $20M NIST allocation shows it gaining concrete traction [9]. The DeepMind Pentagon contract gives the mandatory-vs-voluntary dispute a reference case: voluntary commitments were overridden at the institutional level, by the very organization whose CEO is now championing safety governance.

Demis Hassabis, CEO of Google DeepMind, proposed a U.S. Frontier AI Standards Body modeled on FINRA to conduct independent pre-release testing of frontier AI models for cyber, biological, and deception risks [1][2][3]. The body would operate as a federally overseen public-private partnership funded mainly by industry [4], starting with voluntary lab participation and intending a shift to mandatory compliance — though the trigger, timeline, and decision-maker for that transition are unspecified [1]. The Washington Post editorial board called the proposal 'a good start' [5], and it drew support from Microsoft, Chamath Palihapitiya, and initially OpenAI [6]. The Lawfare proposal by Mark Thomas had already circulated the FINRA model in policy circles before Hassabis's public push [7], suggesting the governance shape had been building toward some form of consensus.

OpenAI has since moved beyond its initial endorsement to propose CAISI — a mandatory federal evaluation framework with NIST as the testing office, annual audits, and incident reporting requirements [8]. The CAISI bill has advanced with a $20M funding plan for NIST [9], and the Institute for Progress has published analysis on its funding requirements [10]. OpenAI has formally diverged from the White House on this design: the Trump administration reportedly favors an approach closer to Hassabis's voluntary model rather than CAISI's mandatory federal authority [11][12]. The result is a genuine three-way split — Hassabis proposes voluntary-then-mandatory industry governance; OpenAI argues for mandatory federal review from the start; and the Trump administration, despite being reported as open to plans similar to Hassabis's [13], is not aligned with OpenAI's more prescriptive design.

The most direct challenge to Hassabis's credibility as a safety proponent comes from Google DeepMind's own conduct. Zvi Mowshowitz's analysis documents that DeepMind signed a Pentagon contract with no restrictions on autonomous weapons or mass surveillance, violating the conditions under which DeepMind was originally sold to Google [14]. An internal petition signed by more than 250 employees and direct lobbying of senior leadership including Jeff Dean and Hassabis failed to stop the deal. Alex Turner resigned from DeepMind, noting that Anthropic maintained its red lines under similar pressure but that pledges of conscience vaporize under institutional pressure [14]. Nate Soares of MIRI added that Hassabis's proposal glosses past expert disagreement spanning a range of roughly 5% to over 50% probability of catastrophic outcomes [14]. The Pentagon contract demonstrates that DeepMind's governance structures have been progressively dismantled since 2014, with the 2018 weapons prohibition removed and the 2026 contract as the endpoint of that erosion.

The structural question none of the proposals' backers has answered is whether any oversight body can credibly issue adverse findings against major labs [1][15]. Harper Foley's legal analysis notes that the federal government's earlier decision to decline pre-release vetting shifted due-diligence burden to enterprise vendor procurement questionnaires rather than eliminating it [16]. The Trump administration's prior non-participation stance now sits against its reported openness to Hassabis's model [13], while OpenAI's divergence from the White House [11] means the regulatory shape is being contested simultaneously in legislative, executive, and industry channels — with the CAISI bill's concrete funding advance the clearest sign that one of those channels is moving.