OpenAI, Anthropic, and Google Collaborate to Establish AI Safety Standards

OpenAI has announced a partnership with Anthropic and Google to enhance AI safety, as reported by Kalshi. This collaboration among three leading AI labs aims to address safety concerns in the development and deployment of AI technologies. The partnership builds on existing efforts such as the Frontier Model Forum, which focuses on creating industry safety standards and evaluations.

The collaboration centers on several key pillars: shared technical evaluations, pre-release audits of advanced models, independent testing frameworks, and standardized safety protocols. Google DeepMind CEO Demis Hassabis proposed the initiative on July 14, 2026, suggesting a US-led standards organization modeled after FINRA, the Financial Industry Regulatory Authority.

OpenAI’s chief scientist Jakub Pachocki reinforced the direction in an essay published September 6, 2026, calling for common safety requirements that would enforce a uniform approach across the entire AI sector. By mid-September, regular working-group meetings were already underway, focusing on the details of shared evaluations and independent audits.

This partnership builds on the foundation laid by the 2024 Frontier Model Forum, which included Microsoft alongside the current trio. All three companies opted out of Nvidia’s Open Secure AI Alliance, which launched in July 2026 and attracted over 30 firms focused on open-source AI security tools.

As these safety standards are established, they may create barriers to entry for competitors, as any firm wishing to secure enterprise contracts or government procurement will need to meet the benchmarks set by OpenAI, Anthropic, and Google.

New Developments in AI Safety Standards

Anthropic, OpenAI, and Google are in discussions to form a standards body focused on the testing and auditing of AI models. This initiative, which began in July 2026, aims to create private mechanisms for ensuring safety and consistency in AI advancements.

Although no formal organization or governance structure has been established yet, the discussions indicate a growing commitment among leading AI firms to collaborate on safety standards. Market reactions suggest a slight increase in confidence regarding Anthropic's valuation, with expectations of achieving certain targets by the end of 2026.

Key indicators to watch include any announcements from the involved companies about formal structures or partnerships, as well as shifts in Anthropic's strategic initiatives that may reflect increased confidence in its valuation trajectory.

New Developments in AI Safety Initiatives

On September 15, 2026, Alex Bores launched "Who Decides," a $30 million organization aimed at creating a unified AI safety agenda for the Democratic Party in preparation for the 2028 presidential election. This initiative follows Bores' previous work as a data scientist at Palantir and his role in co-sponsoring New York's RAISE Act, which is recognized as the strictest state-level AI safety law in the U.S.

"Who Decides" is focusing on building coalitions in 11 strategically selected states: South Carolina, Nevada, New Hampshire, New Mexico, Michigan, Virginia, Arizona, Georgia, North Carolina, Pennsylvania, and Wisconsin. The organization plans to collaborate with labor unions and civil rights groups to develop policy positions and conduct ongoing polling to refine its agenda.

This significant funding reflects the urgency for Democrats to establish a clear stance on AI regulation, especially as tech companies continue to lobby against stringent regulations.

New Developments in AI Safety Legislation

• OpenAI publicly endorsed key provisions of the FRONTIER Act on September 15, 2026.
• The FRONTIER Act, introduced on July 23, 2026, aims to establish a federal standard for AI regulations, replacing state-level frameworks.
• To be regulated under this act, companies must exceed $50 million in revenue and have spent over $1 billion on AI development in the past 36 months.
• The act defines "frontier models" as AI systems requiring more than 10^26 compute operations for training.
• Developers of frontier models will need to publish safety frameworks, conduct annual audits, and report critical incidents to federal authorities.
• Violations of the act could incur civil penalties of up to $10 million per day.
• A new position, Under Secretary for AI Security, will be created within the Department of Commerce to oversee compliance and safety incident tracking.
• The Commerce Secretary will have emergency authority to suspend or restrict AI deployments in cases of imminent catastrophic risk.
• The bill acknowledges state authority over deployment-specific matters and child safety concerns, reflecting industry feedback.