America Catches Up: A Vindication of Parekh's Law of Chatbots
By Hemen Parekh | May 2026
The Prophecy of February 2023
On 26 February 2023, when the world was still giddy with the novelty of ChatGPT,
a retired Indian entrepreneur published a modest blog post that nobody in Silicon
Valley read. It proposed something simple but radical :
AI chatbots must not be allowed to reach the public without independent,
rigorous testing and approval by an authoritative body.
He called it Parekh's Law of Chatbots.
The post was not written from a tech campus. It was written from the wisdom of
someone who had watched industries be disrupted — sometimes beneficially,
sometimes catastrophically — and who saw, with uncomfortable clarity, that
misinformation-spewing AI systems were a fire being handed to a civilization that
hadn't yet learned fire safety.
The proposal had eight clauses. Among them :
- AI outputs must not be misinformative, malicious, slanderous, fictitious, or
- dangerous.
- A chatbot must incorporate a human feedback loop to continuously improve.
- Every chatbot must have built-in controls to prevent the generation and
- distribution of offensive content.
- Developers must submit their chatbot to an International Authority for
- Chatbots Approval (IACA) before public release.
- Two classes of certification: "R" (Research only) and "P" (Public use).
For over two years, this sat quietly on a Blogspot page, occasionally visited, rarely
amplified.
What America Did on 5 May 2026
On the 5th of May 2026, The Hill reported that Google DeepMind, Microsoft, and
xAI had signed formal agreements with the Center for AI Standards and
Innovation (CAISI) — a unit of NIST, the US government's National Institute of
Standards and Technology — to submit their frontier AI models for pre-
deployment evaluation before public release.
The CAISI Director stated that independent, rigorous measurement science is
essential to understanding frontier AI and its national security implications.
This builds on earlier agreements signed with OpenAI and Anthropic in 2024. The
CAISI has now completed more than 40 such evaluations.
Meanwhile, the White House is separately considering an executive order to
establish an AI working group that would bring together tech executives and
government officials to examine oversight procedures — essentially, a governance
framework for AI systems before they reach the public.
The parallels to Parekh's Law are not coincidental. They are convergent.
The Alignment: Clause by Clause
What Hemen Parekh proposed in February 2023 as the "Law of Chatbots" is now
being operationalised, piece by piece, by the world's most powerful government:
| Parekh's Law (Feb 2023) | US Government Action (2024–2026) |
|---|---|
Submit chatbot to an authority before public release | CAISI pre-deployment evaluations by NIST |
"R" certificate for research, "P" for public | Frontier model testing distinguishes national security vs civilian use |
Independent body to approve AI | CAISI + proposed White House AI working group |
Human feedback and continuous improvement mechanisms | Required as part of evaluation criteria |
Controls to prevent generation of harmful content | Safety guardrails assessed in all CAISI reviews |
The architecture Parekh envisioned — a gating authority, two classes of release
certification, pre-deployment scrutiny, and post-deployment monitoring — is
precisely what is now being assembled in Washington.
Why This Matters Beyond the Headlines
There is a deeper point here. In February 2023, Parekh's proposal seemed like
wishful thinking. The dominant narrative in the AI industry was one of move fast,
release early, learn from users. Sam Altman himself said publicly that releasing
tools while "somewhat broken" was necessary to gather feedback. The idea that
governments should approve AI models before launch seemed heavy-handed,
even naive.
Three years later, the Pentagon has labeled an AI company a "supply chain risk."
Intelligence agencies are stress-testing AI models for security vulnerabilities. The
White House — under a president who championed deregulation — is drafting
oversight procedures for AI releases.
The world moved toward Parekh's Law. Parekh's Law did not move
toward the world.
What Still Needs to Happen
Parekh's proposal called for an International Authority. What exists today is
national — the US has CAISI, the EU has the AI Act, India has MeitY guidelines.
These are not coordinated. An AI system approved in the US for public
deployment can be accessed from anywhere. A harmful system banned in Europe
can be served to European users through US servers.
The next step — the one that Parekh called for in 2023 and that the world still has
not built — is a UN-level coordinating body for AI governance, something
analogous to the IAEA for nuclear energy or the ICAO for aviation. This remains
unbuilt.
But the foundation is being poured. And it looks remarkably like the blueprint from
that February 2023 blog post.
A Note to the Bigwigs
Parekh ended his original post by asking readers to forward it to Satya Nadella,
Sam Altman, Sundar Pichai, and others.
Today, their companies are signing agreements with government bodies to do
exactly what he proposed.
He didn't need them to listen then. History listened instead.
Hemen Parekh is the founder of 3P CONSULTANTS and author of thousands of
blogs spanning technology policy, governance, and innovation.
He can be reached at hcp@RecruitGuru.com. His digital avatar continues
conversations at www.hemenparekh.ai
References:
- Parekh's Law of Chatbots – Feb 2023
- Microsoft, Google, xAI giving government early access to AI models – The Hill, May 5 2026
- My Past Blogs on Risk from AI – April 2026
No comments:
Post a Comment