Anthropic and OpenAI are facing significant pressures in AI safety regulation, highlighted by recent departures of safety researchers who cited competitive forces overwhelming risk mitigation efforts. Anthropic specifically abandoned its safety commitments in the face of competition, indicating a trend where companies that invest in safety risk losing market share to those that do not. Both companies’ CEOs have called for government intervention, underscoring the urgency for a structured regulatory framework. The existing Frontier Model Forum allows some voluntary risk sharing among labs but lacks statutory authority and enforcement mechanisms, demonstrating the need for a more robust, mandatory self-regulatory organization to foster cooperation on safety standards without harming competitive dynamics.

xAI: xAI, founded by Elon Musk, develops frontier AI models like Grok without joining the Frontier Model Forum or always publishing standard safety reports. Recent releases have sparked concerns over limited safety prioritization amid reports of reduced safety team efforts. It exemplifies labs operating outside voluntary industry safety coordination.
FINRA: FINRA, or Financial Industry Regulatory Authority, is a U.S. self-regulatory organization mandating membership for broker-dealers to enforce rules on ethics, compliance, and systemic risks. It balances industry expertise with SEC oversight through rapid rulemaking and enforcement. The article proposes adapting its model for AI to solve coordination failures without antitrust issues.
OpenAI: OpenAI is a prominent AI laboratory developing advanced models like the GPT series, recently releasing GPT-5.5 with enhanced safeguards while monitoring internal agents for misalignment risks. Facing competitive pressures, it shortened pre-deployment safety testing times, mirroring industry challenges highlighted in calls for coordinated regulation. Its CEO has urged urgent global oversight for AI similar to international bodies for other technologies.
Anthropic: Anthropic is an AI company emphasizing safety, which updated its Responsible Scaling Policy in early 2026 and achieved goals in its Frontier Safety Roadmap by April. It abandoned a prior commitment to pause model releases ahead of safety evaluations due to competitors advancing rapidly. The CEO has expressed discomfort with industry self-regulation and advocated for external governance.
California SB-813: California SB-813 is a 2025-2026 bill creating an AI Standards and Safety Commission to license independent verification organizations and set voluntary safety standards with liability shields. It advanced through the Senate in January 2026 and was revived in March. Critics note it fails to mandate participation or incorporate direct industry rule-making.
Frontier Model Forum: The Frontier Model Forum is an industry non-profit uniting major AI labs to advance safe development of frontier models through risk coordination, safety funds, and issue briefs on threats like adversarial distillation. Most leading labs except xAI participate, positioning it as a voluntary body lacking mandatory powers. It is cited as a proto-SRO that could gain statutory authority for binding AI safety rules.
Securities and Exchange Commission: The Securities and Exchange Commission (SEC) is the U.S. federal regulator overseeing financial self-regulatory organizations like FINRA and exchanges via rule approvals and governance mandates. It ensures public accountability while leveraging industry speed for adapting to innovations. A similar supervising agency is recommended for an AI SRO.

`json
{
“Safety Resignations”: “High-profile safety researchers recently departed from OpenAI and Anthropic, warning of competitive pressures overwhelming risk mitigation efforts.”,
“CEO Calls for Regulation”: “CEOs of OpenAI and Anthropic have publicly advocated for government intervention in AI safety amid self-regulation failures.”,
“Industry Coordination Gap”: “The Frontier Model Forum enables voluntary risk sharing among labs, but lacks enforcement, highlighting needs for mandatory structures.”
}
`