AI Governance Institute
← News
Corporate Policy2026-08-17

OpenAI Dissolves Preparedness Team, Leaving Frontier Risk Oversight Fragmented

What happened

OpenAI dissolved its preparedness team in late July 2026, according to reporting from the Financial Times covered by The Verge. The team had served as a centralized function responsible for evaluating whether new models crossed serious risk thresholds and for developing mitigation strategies before and after deployment. Those responsibilities have now been distributed across teams organized by specific risk domains, including biology and cybersecurity. This disbandment follows the earlier elimination of OpenAI's AGI readiness and superalignment teams, and comes after OpenAI halted Astra following an internal evaluation that found a critical cyber threshold had been breached, raising questions about whether distributed ownership can maintain the same cross-cutting risk visibility. The cumulative pattern of safety-function changes at OpenAI, combined with the departure of key safety leaders, has drawn scrutiny from researchers and policymakers who see centralized pre-deployment risk assessment as foundational to responsible frontier AI development.

Why it matters

  • ·Enterprises and regulated institutions that rely on OpenAI models now face a materially changed vendor safety posture. The California SB 53 Foundation Model Safety and Security Protocol and similar emerging frameworks explicitly require evidence of a developer's ongoing safety assessment processes, and fragmented ownership may make that evidence harder to obtain and verify through standard vendor due diligence.
  • ·Distributing frontier risk assessment across domain silos creates a structural accountability gap: no single team now holds a cross-cutting view of cumulative or emergent model risks before deployment. Compliance programs that depend on vendor-side safety governance as a first line of defense must now treat that assumption as unvalidated and build compensating controls into their own oversight programs.
  • ·The timing ahead of a potential IPO introduces investor disclosure risk for OpenAI and, indirectly, governance credibility risk for organizations that publicly rely on OpenAI's voluntary safety commitments. Procurement teams that mapped safety obligations to those commitments using controls like PRC-006 or PRC-007 should treat those mappings as requiring immediate re-evaluation.

Governance controls affected

PRC-006Vendor Safety Commitment VerificationPRC-007Vendor Governance Change MonitoringPRC-008Vendor Model Update Disclosure and Re-Assessment ProtocolHOC-001AI System Risk ClassificationBRD-008Voluntary AI Governance Adequacy Standard

What to do now

  • ☐Review your OpenAI vendor due diligence file and identify which safety assurances were predicated on the preparedness team's existence; flag those assurances as unverified until OpenAI publishes updated governance documentation.
  • ☐Assess whether your organization's AI risk classification process for OpenAI-based systems assumed a centralized vendor-side pre-deployment evaluation gate, and determine what compensating internal controls are needed if that gate no longer exists.
  • ☐Update your voluntary AI framework obligation tracker to reflect the changed safety governance structure at OpenAI, and notify relevant business owners who signed off on risk acceptances based on prior commitments.
  • ☐Engage your legal and procurement teams to determine whether OpenAI's restructured safety function triggers re-assessment obligations under existing vendor contracts, particularly any clauses tied to material governance changes.
  • ☐Escalate findings to the board AI risk committee with a clear account of how the change affects your organization's frontier model risk exposure and what additional oversight is now required internally.

What to watch next

Compliance teams should monitor whether OpenAI publishes a replacement governance framework that clarifies accountability for cross-domain risk assessments, and whether the distributed domain teams issue any updated evaluation protocols that can be independently verified. Regulatory bodies in the EU and California are actively scrutinizing foundation model developer governance structures, and the California Transparency in Frontier AI Act may accelerate formal disclosure requirements that would force OpenAI to document its current safety governance structure publicly. The anticipated IPO process will also bring SEC disclosure scrutiny to bear on how OpenAI characterizes its AI safety governance to prospective investors, which could produce new documentation that compliance teams can assess.

Stay ahead of stories like this

Get every US AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Powered by Buttondown.

Related Coverage

Enforcement2026-09-17

Internal Emails Confirm Microsoft and OpenAI Knew Scraping Was Legally Indefensible

Unsealed court filings in the New York Times copyright lawsuit against OpenAI and Microsoft reveal that executives at both companies privately acknowledged that scraping news content for AI training violated fair use principles. A Microsoft director described the practice as potentially the largest theft of labor in human history. The disclosures expose a governance gap between internal risk assessments and continued commercial conduct.

Research2026-09-14

Former OpenAI Safety Staff Signal a Vendor Assurance Gap

A former OpenAI safety employee published an op-ed in the New York Times on September 9, 2026, arguing that competitive pressure is eroding safety governance standards at frontier AI labs. The piece calls on governments to impose clearer safety requirements and slow deployment where necessary. For compliance teams, the primary implication is that relying on vendor self-attestation and voluntary safety commitments may no longer be sufficient as a governance control.

Corporate Policy2026-09-19

Accenture Becomes Anthropic's First Embedded Evaluator, Raising Conflict-of-Interest Questions

Anthropic has announced that Accenture's AI division, Faculty, will be embedded inside the lab to conduct model evaluations, red-teaming, alignment assessments, and safeguard testing. Both companies have committed at least $1 billion over five years. The arrangement is Anthropic's first formal third-party embedded evaluator, but critics question whether a commercially entangled partner can provide genuine independent accountability.