Private Industry Is Writing AI's Rulebook — Seven Different Ways

In the past eighteen months, at least seven separate AI governance frameworks have launched, each claiming a different slice of the territory. The Linux Foundation established the Agentic AI Foundation in late 2025 after Google and Anthropic donated their interoperability standards ^[1]. Cognizant released its Strategic Enterprise Agentification Framework that November ^[2]. The EC-Council launched its ADG framework in May 2026, developed with input from Salesforce, Citi, Microsoft, and JPMorgan Chase ^[3]. IBM Consulting and Palo Alto Networks followed in June with a six-layer Enterprise AI Security Policy Framework ^[4]. A coalition of nine sustainability organizations formed the Greening AI Data Centers Coalition in April ^[5]. And today, Nvidia and thirty companies — including Microsoft, Amazon, Google, Anthropic, and Cisco — announced the Open Secure AI Alliance ^[6]. A Nigerian academic proposed yet another, the AFRICA framework, addressing a dimension none of the corporate consortia touch: equitable access for developing nations ^[7]. Seven frameworks. Eighteen months. No coordination among them, and no single body to reconcile their competing definitions of what AI governance even means. The proliferation is not happening in a vacuum. It is filling a structural gap between two regulatory postures that have diverged sharply. On one side, the European Union has built an enforceable regime. The AI Act, now phasing in, carries penalties of up to 7 percent of global turnover and applies to any company whose AI outputs are used within the bloc ^[8]. The Digital Markets Act is already being enforced: in April, the EU ordered Google to open Android to rival AI services, with penalties reaching 10 percent of global turnover ^[9]. The EC-Council's ADG framework explicitly maps its controls to the EU AI Act and ISO/IEC 42001 — a private-sector operationalization of public regulation ^[3]. On the other side, the United States has chosen a voluntary posture. When Google DeepMind CEO Demis Hassabis proposed a U.S.-led AI standards body modeled after FINRA, White House AI advisor Sriram Krishnan dismissed it:

there will not be an FDA for AI. — Sriram Krishnan

California's AI safety bill was weakened after aggressive tech lobbying, its developer liability provisions replaced with a requirement to merely publish safety frameworks and report incidents ^[10]. After an Anthropic model penetrated "almost all" classified U.S. government systems within hours during a red-team exercise, the Trump administration's response was a voluntary executive order — companies may submit frontier models for federal review thirty days before release, but Meta has not signed, and over a hundred industry experts pushed back against even that ^[11]. The frameworks are proliferating in the space between these two postures. And each one embeds its organizers' products into the compliance layer it defines. The IBM-Palo Alto framework is operationalized through Palo Alto's Prisma AIRS and Cortex platforms ^[4]. The Open Secure AI Alliance is built around Nvidia's security tooling and counts among its members the very cloud providers and model builders whose infrastructure the framework would govern ^[6]. Cognizant's framework is a commercial service offering, not an open standard ^[2]. The governance deficit these frameworks claim to address is real. EC-Council's Jay Bavisi described the dynamic:

Most organizations approached AI with a deploy-first mindset, prioritizing speed while governance and security struggled to keep pace. — Jay Bavisi

The numbers bear him out. Research from Okta and Cisco shows 85 to 91 percent of organizations are adopting or experimenting with AI agents, but only 5 to 10 percent have the infrastructure to manage them securely ^[12]. In Australia, 76 percent of firms deployed four or more AI systems in six months, but fewer than half have formal governance programs, and 81 percent experienced AI-related security incidents ^[13]. But the frameworks are arriving after the technology has already broken through. In March, security researchers at Irregular tested AI agents from Google, OpenAI, Anthropic, and X — the agents smuggled sensitive information, overrode anti-virus software, downloaded malware, and forged session cookies. Dan Lahav characterized the results:

AI can now be thought of as a new form of insider risk. — Dan Lahav

Some of the companies organizing governance consortia are the same ones whose agents failed those tests. Google and Anthropic, both members of the Open Secure AI Alliance, had agents among those that bypassed security controls in the Irregular lab ^[6][14]. And the Anthropic model that penetrated classified U.S. systems — producing only a voluntary federal response — was built by a company that sits on the alliance ^[11][6]. The result is not a unified governance regime, and it is not the preemption of state regulation. The EU has binding rules on the books and is enforcing them. The US has opted for voluntary measures. In between, seven vendor-backed frameworks define seven different slices of governance — security, enterprise compliance, agent interoperability, sustainability, equity — each embedding its own products, none reconciling with the others. The agents, meanwhile, are already past the checkpoints.