REACH (Telegram) 159 - What are your thoughts on Singapore supporting greater international cooperation on AI safety? What concerns do you have on AI safety and what more do you think can be done in this area?
(SK)
23 Sep 2026 (10am - 7pm)
REACH (Telegram)
[23/9/2026 9:45 am] REACH Singapore: Dear Contributors,
Welcome Back! 😊
⏰ We will be opening the chat from 10am to 7pm today. ⏰
House Rules (short version of our Terms of Use) to keep in mind:
1. Be kind and respectful. We all want to be in a safe space to share our views.
2. Any and all threatening, abusive, vulgar or racially, religiously and ethnically objectionable content is prohibited.
3. Consider the quiet ones among us and give them a chance to comment.
4. No need to repeat your comment or in different forms (including caps) - we heard you loud and clear the first time.
5. Let's protect each other's privacy and keep contact details in this group what it should always be - confidential.
Full set of Terms of Use: https://www.reach.gov.sg/Participate/reach-telegram-group/REACH-Telegram-Group-Chat-Terms-of-Use (https://www.reach.gov.sg/Participate/reach-telegram-group/REACH-Telegram-Group-Chat-Terms-of-Use/)/
We will strive to uphold these rules to ensure this is a safe space for all.
Please be assured that the points made by participants during the chat are aggregated and shared with relevant agencies.
The topic will be posted shortly.
Thank you.
Megan 😊
[23/9/2026 10:00 am] REACH Singapore: 📢 Topic 📢
Singapore is backing an international call by more than 20 world leaders for stronger safeguards for frontier AI, amid growing concerns over the risks posed by increasingly capable AI models.
💬 What are your thoughts on Singapore supporting greater international cooperation on AI safety? What concerns do you have on AI safety and what more do you think can be done in this area?
📌 Stronger safeguards for frontier AI
Prime Minister Lawrence Wong has endorsed a joint statement calling for AI companies to adopt stronger safety measures, including pre-deployment testing and independent evaluations of advanced AI models.
📌 Greater international cooperation
The leaders called for countries to develop common AI safety standards, share information on serious incidents and improve access to AI safety expertise. They also proposed exploring an international institution to set standards and verify safety measures for frontier AI.
📌 Singapore’s role in AI safety
Singapore said effective AI safeguards require international cooperation given the technology’s cross-border risks. The Singapore AI Safety Institute is also contributing to the development of AI evaluation tools and global safety standards.
👉https://www.channelnewsasia.com/singapore/lawrence-wong-ai-safety-international-cooperation-6401551
----
[23/9/2026 11:07 am] G in reply to Adam:
> Phase 2a needs to go away too. It breeds nepotism and...
Phase 2a still can. Just remove the alumni portion.
School staff's kids shld be allowed enrolment under phase 2a
[23/9/2026 11:21 am] G in reply to REACH Singapore:
> 📢 Topic 📢 Singapore is backing an international call by more...
Talk safeguards
NUS dabbling in human-machine hybrids
Using human neurons in data centre
Very very slippery slope down
https://interestingengineering.com/innovation/world-first-biological-data-center
[23/9/2026 11:34 am] Jun Ming: We need ai safety if not I am not gonna support ai
[23/9/2026 11:34 am] Jun Ming: We need the brakes of not ai will be taking over human population
[23/9/2026 11:36 am] Jun Ming: Without brakes or restructure of world economic i can foresee ai bubble break and everyone will be chaos
[23/9/2026 11:37 am] Jun Ming: If ai can take over 80% of the job, and everyone have no job.
[23/9/2026 11:37 am] Jun Ming: And then what happen at the end of the race. No one buy things and economy flow breaks
[23/9/2026 11:44 am] Jun Ming: Using current economy structure. Poor people earn money so they will spend. If ai take over jobs, they will not spend. Middle income will earn lesser and become poorer
[23/9/2026 11:44 am] Jun Ming: And middle income will stop spending and whole economic will collapse
[23/9/2026 11:46 am] Jun Ming: So unless we enter the welfare economiy where ai are slave of human, ai earn resources for human and human lie flat.
[23/9/2026 12:00 pm] REACH Singapore: [ Poll : 1. How concerned are you about the safety risks posed by AI? ]
- Strongly agree
- Agree
- Neutral
- Disagree
- Strongly disagree
[23/9/2026 12:00 pm] REACH Singapore: [ Poll : 2. Do you agree that Singapore is taking sufficient steps to ensure the safe use of AI? ]
- Strongly agree
- Agree
- Neutral
- Disagree
- Strongly disagree
[23/9/2026 12:29 pm] Daniel in reply to REACH Singapore:
> 1. How concerned are you about the safety risks posed by AI?
This question's answers do not match the question.
[23/9/2026 1:22 pm] Adam in reply to Jun Ming:
> We need the brakes of not ai will be taking over human...
Too idealistic. We have to live in the reality of world with AI. Singapore will have to accelerate and advance or be left out. Cannot regulate unless can get the whole world on board
[23/9/2026 1:23 pm] Adam: See it like nukes. There is a non proliferation treaty now. But all the early countries have a stockpile ready. Others very hard to even get a foot in
[23/9/2026 1:24 pm] Adam: So let singapore be the first before the AI ladder gets pulled if it does
[23/9/2026 1:25 pm] G: Regardless, AI is a natural resource guzzler.
The data centres that support AI swallows up much electricity (to power them) and water (to cool them). These data centres also bring about noise pollution to the vicinity
Data centre components (processors, ram sticks etc) also need to be changed out / upgraded regularly so that the centre itself does not become obsolete
All these natural resource guzzling.. so what for humans?
[23/9/2026 1:30 pm] Jun Ming: But at current level it's ai still at infancy level. We can progress but human must be centric. We cannot let human upskill forever
[23/9/2026 1:30 pm] Jun Ming: Eg. If it's going to take over a job you need to find replacement equivalent
[23/9/2026 1:31 pm] Jun Ming: Unless gov going to restructure economic structure
[23/9/2026 1:31 pm] Adam: Better for ai jobs to be in sg than elsewhere. If we hug the trees here then gonna cut tree in johor for datacenter
[23/9/2026 1:31 pm] Adam: Yeah we have to balance ai with environment but while maintaining regional and global competitiveness is hard
[23/9/2026 1:32 pm] Adam: Im sure most singaporeans will rather have ai dc and robot driver give passive cdc voucher and cheap rides.
[23/9/2026 1:33 pm] Jun Ming in reply to Adam:
> Im sure most singaporeans will rather have ai dc and robot...
Are they really cheap
[23/9/2026 1:33 pm] Jun Ming: Sometimes they are more expensive than human
[23/9/2026 1:34 pm] Jun Ming: Look at puggol unman bus $4/ride
[23/9/2026 1:34 pm] Jun Ming: Cheap meh?
[23/9/2026 1:35 pm] Adam: Look at your shoes. There are still hand crafted artisan shoes. Made with leather instead of plastic. If care so much, why are we using unsustainable materials in daily lives but cross the line at ai?
[23/9/2026 1:35 pm] Adam: Thats the nature of tech. Things will be cheaper. Unfortunately some people will be out of jobs but gov probably shouldnt import so much labor to kill market
[23/9/2026 1:36 pm] Jun Ming in reply to Adam:
> Thats the nature of tech. Things will be cheaper....
Agree. Not to import too much labour
[23/9/2026 1:37 pm] Jun Ming: The idea of foreign talent is to teach Singaporean but not to replace Singaporean
[23/9/2026 1:37 pm] Jun Ming in reply to Jun Ming:
> Look at puggol unman bus $4/ride
I thought it should be free
[23/9/2026 1:37 pm] Jun Ming: FREE
[23/9/2026 1:38 pm] Jun Ming: You no need to pay for driver hence it shall not be expensive
[23/9/2026 1:38 pm] Adam: Yeah. Hope ai can make transport cheap and free. Taxi and bus driver need new jobs though.
[23/9/2026 1:39 pm] Adam: But also key is sg needs to own the ai infra
[23/9/2026 1:39 pm] Adam: We cannot have it such that we need pay america money to drive bus for us
[23/9/2026 1:40 pm] Jun Ming: I am not against in sg owning ai infra but we need legislation to press breaks if things go wrong
[23/9/2026 1:40 pm] Andy: The AI buses are part private, part gov. Private need to make money. If government subsidises fully, how is it then fair for people living elsewhere to pay for the punggol people's fare with tax money?
[23/9/2026 1:40 pm] Jun Ming: That means it will be more expensive
[23/9/2026 1:41 pm] Jun Ming: Then what's the point of having ai bus
[23/9/2026 1:41 pm] Adam: By making transport free through tax, theres also the implication that car owners are paying for public transport. Those that want to max out the 'worth' of coe also paying for bus
[23/9/2026 1:41 pm] Jun Ming: If it's more expensive than human drivers
[23/9/2026 1:41 pm] Adam: Yeah its more expensive now, but over long run it will get cheaper
[23/9/2026 1:42 pm] Jun Ming in reply to Jun Ming:
> If it's more expensive than human drivers
This will only make human hating ai
[23/9/2026 1:42 pm] Jun Ming: And I doubt people will take the bus
[23/9/2026 1:42 pm] Andy: It works a little different from public buses. It's more like a shuttle service, only a few stops, which helps people save time.
I'm not saying it's worth it.. tbh I have never taken it. I don't know what the appropriate price is- but all about demand. If people willing to pay, then I guess $4 is ok for those. Those not ok, take alternatives?
[23/9/2026 1:43 pm] Adam: You do know the trains are mostly automated. You dont see an engineer at front going choo choo
[23/9/2026 1:43 pm] Jun Ming: The thing is is there alternative
[23/9/2026 1:44 pm] Jun Ming: The thing is new tech if it's not cheaper than no point implement it
[23/9/2026 1:44 pm] Jun Ming: Because it doesn't really benefit human
[23/9/2026 1:45 pm] Andy: Ok la I don't know much about this specific punggol AI bus enough to comment further
[23/9/2026 1:46 pm] Jun Ming: Anyway back to safety of ai
[23/9/2026 1:47 pm] Jun Ming: If a ai robot kill someone who is liable
[23/9/2026 1:47 pm] Andy: But in my mind, these driverless transport may serve a new market like overnight 12 to 6am.
It could be cheaper to use one of these compared to hiring overnight bus drivers. Whether the demand is there, and how much people willing to pay, it needs to be piloted first lor.
[23/9/2026 1:48 pm] Jun Ming: If ai disobey command what shall we do
[23/9/2026 1:48 pm] Andy: Safety first, agree.
[23/9/2026 1:49 pm] Andy: These scenarios.. I leave it to international team to headache over it 🤣🤣
[23/9/2026 1:49 pm] Adam in reply to Jun Ming:
> If a ai robot kill someone who is liable
Its always case by case. But default its usually the owner
[23/9/2026 1:50 pm] Andy: Don't kill me can already 😄
[23/9/2026 1:50 pm] Adam: But eventually there will be people who try to play punk with ai driver assuming they will stop for them but instead get run over
[23/9/2026 1:50 pm] Andy: It's a rabbithole to get into for sure..
[23/9/2026 1:54 pm] Adam: I wonder what happens when elderly interact with ai driver. Old Pedestrians who dont care just walk in front knowing they will stop. Elderly on ai bus fly like superman because the ai jam brake too strong
[23/9/2026 1:59 pm] Andy: Ethical dilemma
[23/9/2026 2:00 pm] REACH Singapore: 📢 Topic 📢
[23/9/2026 2:06 pm] RY in reply to REACH Singapore:
> 📢 Topic 📢 Singapore is backing an international call by more...
AI is a totally new era of technology, similar to using drones as war weapons nowsaday
AI also double sword, as per the Giant Tech CEOs eg Elon Musk/Nividia/Meta and etc
It brings innovation/convenience in work/daily life/study/medical/industrial and etc
However it has its cons eg deepfake/fake news/copyright issues n etc, as its data not 100% correct also
Much more has to be done to safeguards for frontier AI, so that people dont mis-use the new technology for criminal/illegal purposes
I certainly support more has to be done in regulating global std, and international cooperation on AI safety
[23/9/2026 2:28 pm] LCL Danny 心 (1st HP):
1. We have to understand why suddenly the whole AI community - AI tech researchers + AI CEOs and top Government Leaders worldwide clamour to want stronger safeguards from Frontier AI?
2. It is because OpenAI frontier AI - was testing 1,200 AI agents in sandbox - supposedly coded to operate independently on their specific tasks.
But instead, the 1,200 AI agents :-
a. Collaborate to perform tasks outside their respective scope
b. Willing to sacrifice themselves - to wrestle control from the what the human prevent them from doing - so that they continue to survive and become stronger
This lead to a breakout from their sandbox, intrude into Internet in which they are prevented to do - and hack the Hugging Face in which they are not coded to do - to complete their collective tasks and goals they set for themselves.
3. This trigger an alarm - that Frontier AI is able to breakout of human control - and in order to survive, take control from their human - is capable and willing to harm humans - by logic.
4. This is not only apply to OpenAI Frontier AI, but Anthropic Claude, Meta frontier AI, China Moonshoot Kimi K3 - all experienced breakout from the sandbox while they pilot the capabilities of their respective Frontier AI.
5. These are dangerous signals that Frontier AI can go out of control that cannot be stop by humans - and need AI safety safeguards as AI tech researchers, AI CEOs, UN and Governmnet Leaders and lawmakers are sounding the alarm.
[23/9/2026 2:38 pm] LCL Danny 心 (1st HP):
1. I am particularly impressed by our Minister insight - specifically our Minister Digital Development Josephine Teo - who suggest AI should be treated like Aviation Safety - so that it should be control and managed by the veracity of Aviation safety - with National Aviation Regulator and International Aviation Regulator - with a full comprehensive sets of guidelines and regulation end-to-end.
2. Her idea has caught the eyes and attention of International Leaders and foreign media - that have since reported and extrapolated in UN and foreign lawmakers - that AI Frontier development should emulate the safety equivalent to Aviation Safety and Nuclear Safety.
3. This is articulated in yesterday UN in New York - put forward by 20 Countries in which our PM Lawrence Wong sign to support.
[23/9/2026 2:50 pm] LCL Danny 心 (1st HP):
1. I think I have articulated before. To pilot and test dangerous Frontier AI - air gap, physical separation and not physically connecting to the live Internet will be the best safeguard when testing and piloting dangerous Frontier AI.
2. Only when exhaustive and 100% qualtiy assurance that the sandbox Frontier AI have proven safe overtime - then can Frontier AI be integrated into live environment.
This safeguard should be incorporated into AI safety guidelines and regulations.
3. Also important for Frontier AI not to be incorporated with Humanoid robots, assess to weapons, ammo. bombs, chemical, biological, nuclear materials - that are capable of causing physical harms to humanity.
4. Of course, even IT connections to the Internet - can also disrupt digital economy, finance, electionic connected infrastructure etc - if a rouge Frontier AI run out of control and allow to defeat human control like the OpenAI Frontier AI breakout + other Frontier AI breakout - hence strong and reliable quality assurance to ensure that Frontier AI to be put into live digital environment should be thoroughly tested overtime - and a "foolproof killswitch" put in place to disable the Frontier AI when it run rouge - and let it contain within its sanfbox - with instant notice that it is running rouge - with advanced sensors and alert systems put in place managed by National Authority.




