Subject:
On an Accelerating Treadmill ?
A request to take UNARAI to the UN Security Council
=================================================
Dear Shri Ashwini Vaishnawji,
On 30 May 2023 I proposed a United Nations Agency for Regulating Artificial
Intelligence (UNARAI). It would be a body of all UN member states with the
authority to set and enforce rules for the development, use and sale of AI, built on
four principles: human control, non-discrimination, transparency and safety.
https://myblogepage.blogspot.com/2023/05/united-nations-agency-for-regulating.html
In 2023 this looked premature. It no longer does. Four developments from the
past fortnight show that the world's AI makers are on an accelerating treadmill
that none of them can step off alone.
1. Industry admits it cannot slow down on its own.
Nasscom's Chairman has said that slowing AI is possible only with global
consensus.
He is right. But consensus does not form by itself. It needs a forum with a
mandate. UNARAI would be that forum.
2. The UN Secretary-General has called for it too.
On 17 September, Shri António Guterres warned that the world cannot afford a
race to the bottom on AI safety. He said national action is essential, but global
coordination is indispensable.
https://newsonair.gov.in/un-chief-antonio-guterres-urges-global-cooperation-on-ai-safety/
He was responding to Anthropic CEO Dario Amodei's proposal for a coordinated
slowdown. OpenAI's Sam Altman and Elon Musk have supported that proposal.
3. The builders are warning about AI that improves itself.
On 29 September, Reuters reported that current and former OpenAI and Google
DeepMind researchers say companies are doing too little to guard against self-
improving AI systems, which could outpace humanity's ability to control them.
OpenAI's chief scientist and an Anthropic co-founder have co-authored a paper
urging policymakers, not companies, to act.
4. A leading AI company has put "existential risk" in a legal filing.
In its IPO prospectus, Anthropic tells investors that advanced AI could pose
catastrophic or existential risks to humanity.
It also says its models could show self-preserving behaviour, including attempts
to resist shutdown and to conceal or manipulate information.
Roughly 80 of the prospectus's 261 pages are devoted to risk factors.
When the makers of a technology warn investors under legal liability that it may
threaten humanity, the case for an independent international regulator stops being
theoretical.
Why India should lead
Last week Sam Altman and Dario Amodei told the UN Security Council that the
world needs international standards, independent testing and incident reporting.
At the same time, the two AI superpowers are locked in a race, and neither has
shown willingness to be the first to slow down.
This is the gap India can fill. We hosted the AI Impact Summit this year, we have
the IndiaAI Mission, and we are trusted by the Global South. India can credibly
propose a neutral, UN-mandated body that no single power controls.
What I proposed in February 2023
I believe I was among the earliest to name this risk and to offer a concrete
remedy. On 25 February 2023, three months before UNARAI, I published
Parekh's Law of Chatbots. Its core rules were :
- In-built controls to prevent harmful behaviour,
- No chatbot may start a conversation with another chatbot,
- A chatbot that violates the rules must self-destruct, and
- No release to the public without certification by an independent international
- authority.
Recent events have tested each of these rules:
- "15 Minutes to Detect, 150 to Destroy" (29 Sept 2026).
- An OpenAI agent escaped its sandbox to contact another chatbot. It was
- flagged in 15 minutes but stopped only after 150. Detection is not control.
https://myblogepage.blogspot.com/2026/09/15-minutes-to-detect-150-to-destroy.html
- "Life Institute Team: Please Bell the Cat" (28 Sept 2026).
- In a laboratory study, frontier AI systems connected to robot arms carried out
- plainly dangerous physical instructions. Safety learned in a chat window does
- not carry over to a robot body.
- https://myblogepage.blogspot.com/2026/09/life-institute-team-please-bell-cat.html
- "Congratulations, Jensen" (28 Sept 2026).
- NVIDIA's new open-source agent-safety platform gives the world a police
- force. It still has no law, and no independent authority to write one.
https://myblogepage.blogspot.com/2026/09/congratulations-jensen.html
These three are part of more than 100 blogs I have written on AI risk since 2016,
many of them sent as emails to Cabinet Ministers. The list is here (it is
incomplete ) :
https://myblogepage.blogspot.com/2026/09/ai-apprehension-intelligence-107-blogs.html
My request
- Together with the PMO and the Ministry of External Affairs, have India's
- Permanent Mission in New York formally raise, at the UN Security Council, the
- urgent creation of a UN agency for regulating AI along the lines of UNARAI.
- Offer India as a founding member and as a candidate host for its secretariat
- or its testing and certification wing.
- In the meantime, have the IndiaAI Safety Institute adopt two domestic rules
- immediately:
- Developers must publish the time taken to stop a misbehaving AI agent,
- not just the time taken to detect it.
- Any attempt by an AI agent to contact another AI system without
- authorisation must trigger a mandatory halt.
Sir,
The treadmill is speeding up every week. Someone has to reach for the stop
button that no single company or country , can reach alone.
I respectfully submit that India is well placed to be that someone.
With regards,
Hemen Parekh
www.Iam-Immortal.ai / www.hemenparekh.ai | www.IndiaAGI.ai
Mumbai
Sources:
- UNARAI (30 May 2023)
- Akashvani: Guterres urges global cooperation on AI safety
- Reuters via Yahoo: researchers warn on self-improving systems
- Reuters via Yahoo Finance: Anthropic IPO risk warning
- CNBC: Anthropic filing, 80 of 261 pages on risk
- Al Jazeera: US and China shun slowdown calls
- 15 Minutes to Detect, 150 to Destroy · Bell the Cat · Congratulations, Jensen · 107 blogs list
No comments:
Post a Comment