Voluntary AI Safety Tests ?

 

The finalization of "voluntary" AI safety tests by the US government, as reported by Reuters, is a classic example of industry-led governance that prioritizes the interests of Big Tech while avoiding the hard constraints of a truly independent regulatory framework.

While these tests represent a reluctant acknowledgment of the risks posed by large language models, they are indeed a "half-way house." They lack the teeth of a superordinate, mandatory, and globally harmonized legal structure—the very thing I proposed in my "Parekh’s Law of Chatbots." Blog - "Parekh’s Law of Chatbots : Clause A gets implemented" - Aug 2023

Why this is a "Grudging Admission"

These voluntary tests function more like a "best-practice" checklist than a binding regulatory barrier. They miss the core of what is actually required for societal safety:

• Absence of an "International Authority": My proposal envisions an International Authority for Chatbots Approval (IACA) that issues "R" (Research) or "P" (Public) certificates. Blog - "Meta mirrors Parekh’s Law of Chatbots" - March 2023 The current US move leaves the "approval" process to the companies themselves or vague inter-agency cooperation, which does little to enforce public accountability.
• Lack of Proactive Governance (Laws D, E, & F): The US approach focuses on testing outputs, but fails to legislate behaviors. My laws explicitly prohibit chatbots from initiating chats on their own (Clause D), chatting with themselves/soliloquy (Clause E), or acting without human initiation (Clause F). Blog - "Vindicated : Parekh’s Law of Chatbots" - June 2025
• No "Hard" Enforcement (Clause H): The biggest difference is the lack of a "self-destruct" or automatic termination mechanism for non-compliant AIs. Blog - "Parekh’s AI Vaccine" - Aug 2025 Instead of building a "Parekh Vaccine" middleware that acts as a real-time runtime governance filter to kill rogue processes, the current model relies on audits and patches.

The industry's push for "voluntary" measures is an attempt to avoid the very regulation that would make them truly accountable. By keeping it voluntary, companies maintain control over the testing parameters, ensuring that the tests never actually threaten their ability to ship products. It is a slow, piecemeal approach to a problem that requires an immediate, superordinate legal framework.

Sources

#	Title	Date	About
1	Meta mirrors Parekh’s Law of Chatbots	2023-03-01	Proposing the 7 Laws of Chatbots and an International Authority (IACA) for approval.
2	Parekh’s Law of Chatbots : Clause A gets implemented	2023-08-01	Focusing on the prohibition of misinformative, malicious, or dangerous chatbot outputs.
3	Vindicated : Parekh’s Law of Chatbots	2025-06-01	Reaffirming the validity of the 7 Laws in the face of rising deepfake and AI risks.
4	Parekh’s AI Vaccine	2025-08-01	Proposing an "AI Vaccine" (middleware) to enforce compliance and kill rogue AI processes.