Vonghour Leng

Some conversations stay with you long after the room has emptied.

Last week at the Rencontres Économiques d'Aix-en-Provence, I had the privilege of presenting alongside some of the most forward-thinking minds in global economics and geopolitics.

My sincere gratitude to Jean-Hervé Lorenzi and the organising team for their invitation — and for creating a space where ideas truly matter.

I also want to take a moment to recognise two exceptional leaders from The Asian Network , whose presence made these exchanges all the more meaningful:

🇰🇭 H.E. Ing Kantha Phavi, Minister of Women's Affairs of Cambodia — a powerful voice for gender inclusion, human dignity and sustainable development across Southeast Asia. Her leadership is a reminder of what principled governance looks like.

🇯🇵 Prof. Akiko Yamanaka, Former Vice-Minister for Foreign Affairs of Japan — a distinguished diplomat and intellectual whose commitment to multilateralism and regional cooperation continues to shape the future of Asia-Pacific relations.

This is what The Asian Network is built for: bringing together leaders across borders, institutions and perspectives to forge the partnerships that matter.

Asia is asserting itself as the world's new center of gravity. But its rise is built on fault lines — demographic, climatic, geopolitical — that demand exactly this kind of dialogue.

Missed the sessions or want to revisit the debates? All talks are available to watch in full 👇
🎥 lnkd.in/eXTaPT7F

And thanks to our commenter, NGO Ravindra - who explores all these paradoxes in depth in my latest piece for the World Economic Forum 👇
🔗 lnkd.in/ei4SGgKi

Catching-Up With ‪@vonghourleng‬ On:
- Medium: medium.com/@vonghourleng
- Substack: substack.com/@vonghourleng
- YouTube: ‪‪@vonghourleng‬
- TikTok: www.tiktok.com/@vonghourr
- Facebook Page: www.facebook.com/criticalbyvonghour

2 weeks ago | [YT] | 0

Vonghour Leng

How can #AI serve people, rights, and the public good? At @UNESCO, we are helping countries turn AI ethics into action.

Ahead of the first UN Global Dialogue on AI Governance and the AI for Good Summit in Geneva (6–10 July), UNESCO’s resources and tools can help shape the conversation — and help governments build an AI #ForThePeople.

This is UNESCO in action: advancing human rights, building trust, and helping countries shape AI strategies that serve everyone.

Learn more here: unesco.org/ethics-ai/en

Video Funded: @EU_Partnerships.
Video Source: www.facebook.com/share/v/1BHDrY1Uqc/?mibextid=wwXI…


Affiliated With: ‪@unitednations‬, ‪@UNESCO‬, ‪@InternationalPartnershipsEU‬, ‪@UnitarOrg‬, ‪@UNIDIR‬, ‪@UNDP‬, ‪@UNIDOofficial‬, ‪@unescap‬

2 weeks ago | [YT] | 0

Vonghour Leng

Blog Post II: Reflection on Global Conference on AI, Security and Ethics 2026 (AISE26)

The Unseen Architects: 5 Impactful Takeaways from the UNIDIR Global Conference on AI, Security and Ethics

1. Introduction: The Speed of Innovation vs. The Weight of Responsibility

In Silicon Valley, the dominant religion is "996" — a grueling culture where developers grind from 9 AM to 9 PM, six days a week. As Chris Shank, founder of Simplex Heritage, noted at the UNIDIR Global Conference on AI, Security and Ethics 2026 (AISE26), this manic velocity has created a world where innovation moves at "machine speed." Meanwhile, the diplomatic frameworks designed to keep us safe are still effectively "booting up."

The friction is visceral. We are witnessing a clash between the hyper-acceleration of code and the deliberate, necessary slowness of international responsibility. The summit in Geneva revealed that while the technology is ready, our governance is struggling to clear the BIOS. The following five takeaways highlight a world where global security is no longer just the province of generals, but is being quietly reconfigured by investors, coders, and the fragile digital realities of the global majority.

2. Takeaway 1: The "Gordian Knot" of Responsible Capital

The traditional firewall between "civilian" and "combatant" technology has dissolved. Historically, ethical investors relied on simple Environmental, Social, and Governance (ESG) metrics: if a company made more than 10% of its revenue from weapons, you cut them off. In the age of dual-use AI, that metric is a relic.

The Failure of Revenue-Based Exclusions — Sam Jones, President of the Hardland Initiative, described a new, terrifying reality for the financial sector. A startup building a civilian API or a cloud provider hosting medical data can be repurposed into a military targeting system in a heartbeat. This creates what Jones called a "Gordian knot of risk" for investors trying to fulfill ethical mandates.

The Data Vacuum — The strategic risk is that investors are now the de facto gatekeepers of global security, yet they are operating in a data vacuum. Traditional ESG providers aren't offering "fit-for-purpose" analysis for AI. The shift must move from "revenue-based exclusions" to "product and conduct-based assessments" — monitoring not just what a company sells, but how its code is actually behaving in conflict zones.

3. Takeaway 2: "Collateral in the Code" — The Small Island Vulnerability

The emotional and strategic center of the summit was the testimony of Ashley Robinson, Chair of Policy for the Jamaica Artificial Intelligence Association. Her argument was simple: Small Island Developing States (SIDS) are becoming "collateral in the code" — downstream victims of AI failures they didn't design and cannot control.

The Connectivity Paradox — Islands like Jamaica face a "connectivity paradox." They have high digital engagement but sit on "fragile infrastructure" owned by foreign private entities. When a primary network fails, these nations must reach for private satellite constellations that operate outside of local treaties. Sovereignty, in these moments, is held hostage by a private decision chain an ocean away.

Inclusion as a Security Mandate — For a tourism-dependent nation, a modeling error isn't a glitch; it is an existential threat. If an AI targeting system trained on Northern data misclassifies a cruise ship or a fishing fleet, it triggers a national economic catastrophe. Robinson’s message to the diplomatic community was a piercing correction: "You can not capacity build your way out of exclusion. The remedy for exclusion is not capacity; it is inclusion."

4. Takeaway 3: The "Rubber Stamp" Trap and the Hollowing of Human Control

We talk incessantly about the "Human in the Loop," but Amy Chang of Cisco presented the "subversive trilemma": the reality that speed, intensity, and control are in a constant, losing trade-off.

The Human Bottleneck — In the chaos of "machine-speed" warfare, human judgment is no longer a safeguard; it is a bottleneck. When an AI system flags 23,000 vulnerabilities across 10,000 projects in a single month, a human operator can not "verify" the data. They can only react. Under this pressure, meaningful oversight is hollowed out, turning the operator into a mere "rubber stamp" for an automated system they don’t fully understand.

Erosion of Accountability — This "deference to technology" creates a legal fiction. If a commander clicks "accept" on a machine-generated target because the machine is too fast to interrogate, the chain of command responsibility — the very foundation of international law collapses.

5. Takeaway 4: Attacking Human Dignity — The Threat to the Geneva Law

Global debate usually focuses on the "Hague Law" (how we kill). But Matthew Kennedy and Nathan Heath shifted the focus to the "Geneva Law" (how we protect the vulnerable). Their findings on deepfakes and Prisoners of War (POWs) were chilling.

The Liar’s Dividend — Generative AI (GenAI) allows malicious actors to clone the voices of captured soldiers or fabricate "capture cards" with false health data. This doesn't just hurt the individual soldier; it attacks the entire information system of the Geneva Conventions. When the ICRC or a family can no longer trust a video of a loved one, we enter what Kennedy and Heath called "this black darkness of not knowing."

Strategic Dehumanisation — The "Liar’s Dividend" means that the mere possibility of a deepfake degrades the validity of all legitimate human rights reporting. Human dignity is just as vulnerable to an algorithm as a tank is to a drone.

6. Takeaway 5: The Fragile Bridge — A Rare Moment of US-China Consensus

In a landscape of fractured geopolitics, David Edelman and Xiaosen detailed a significant breakthrough: a "track two" dialogue between the US and China that survived even the lowest points of their bilateral relationship.

Consensus Born of Foreboding — The two superpowers reached a rare, revolutionary agreement: humans must maintain control of nuclear launch decisions. This was not born of mutual trust, but of mutual "foreboding." Both sides recognise that in an AI arms race, a machine-driven nuclear escalation is a catastrophe with no winner. This "no-brainer" agreement is a vital starting point for global norms, proving that AI is forcing even the most bitter rivals back to the table out of shared existential fear.

7. Conclusion: The Agile Path Forward

As the summit closed, Ambassador Philip Tigo offered a final, strategic distinction. AI governance is not a "complicated" problem — one that you solve once, like building a jet engine. It is a "complex" system — one that requires constant vigilance, like an ecosystem.

We can not rely on static, "one-and-done" regulations. Governance must be as agile as the code it seeks to restrain. It must include the voices of those on the "collateral surface" of the technology. If we continue to prioritise speed over safety, we are not just building better systems; we are building a more efficient collapse.

A Final Thought: If the algorithm is moving at the speed of light, are we willing to be the ones who slow it down to save ourselves?



More On:
- Day 1 of the UNIDIR Global Conference on AI, Security and Ethics 2026.
youtube.com/live/MVD7T0vGw7Y?si=tWC4jeM5fTD6q5vL
- Day 2 of the UNIDIR Global Conference on AI, Security and Ethics 2026.
youtube.com/live/Cf8-MO7DYlE?si=9tnn39BHtnBACwN6
- Vonghour Leng’s Playlist on UNIDIR Global Conference on AI, Security and Ethics 2026.
youtube.com/playlist?list=PLt...

3 weeks ago (edited) | [YT] | 0

Vonghour Leng

Blog Post I: Reflection on Global Conference on AI, Security and Ethics 2026 (AISE26)

The Ghost in the Command Chain: 5 Surprising Realities of AI Warfare in 2026

At a high-level security summit in early 2026, Kristen Binar of the Johns Hopkins University Applied Physics Lab posed a deceptively simple question: "have you ever been in a meeting where everyone seems to agree, only to realise later that everyone understood the core terms differently?" This "lost in translation" phenomenon is no longer just a boardroom friction. In the theater of modern conflict, this semantic misalignment has become a primary vulnerability, potentially shattering our existing governance models for military Artificial Intelligence (AI).

The UNIDIR 2026 Global Conference on AI, Security and Ethics in Geneva has made one reality chillingly clear: the debate is no longer about if AI will shape security, but how it is already doing so. With sensor-to-shooter times collapsing to mere seconds and AI-enabled targeting support becoming standard in active conflicts, the pace of battle has outrun our ability to monitor it. We are witnessing a shift where the AI industry’s relevance has moved from the commercial periphery to the strategic center, turning data centers and cloud services into objects of geopolitical contestation.

To survive this high-rapidly shift, we must move beyond abstract ethics and confront the operational friction where code meets the kinetic. The following five counter-intuitive realities challenge the conventional wisdom of military AI, revealing how technology is externalizing sovereign authority and fundamentally altering the human role in the kill chain.

1. The Semantic Trap: Bias is "Math" to Engineers But "Justice" to Policymakers

A dangerous gap exists between those who build AI and those who govern it. When stakeholders commit to "mitigating bias," they are often moving toward irreconcilable goals. For an engineer, bias is a technical property — a statistical distance from a "ground truth." For a policymaker, bias is a value-based property — a distance from equity, justice and transparency, or International Humanitarian Law (IHL). This means a system can be statistically "accurate" while remaining socially or legally unacceptable.

The challenge for the policymaker is compounded by three distinct sources of bias. While an engineer can attempt to "fix" computational bias (errors in sampling or measurement), they cannot simply "fix" systemic bias, which is often a faithful encoding of a flawed reality. When history and institutional inequities are baked into the data, the model isn't "broken" — it is reflecting a slice of reality shaped by human collection options (human behaviour and/or desired interests). Finally, human cognitive bias (such as automation bias) ensures that even a "perfect" model can produce a catastrophic outcome if the operator over-trusts the machine’s output.

"To Jane [Pinellis], a biased outcome is about distance from statistical truth and to me a biased outcome is about distance from justice. Same word but one points to math and the other one to our values." — Kristen Binar.

2. The Proactivity Paradox: How AI Narrows the "Cognitive Spotlight"

We are transitioning from reactive tools to "agentic" proactive systems. While proactive AI can reduce information overload, it creates a "proactivity paradox" that narrows a commander’s agency through a "cognitive spotlight" effect. When a machine prioritises specific threats or zones, the operator’s attention naturally follows, rendering anything outside that machine-generated focus effectively invisible. This creates "responsibility creep," where the machine pre-frames the problem so thoroughly that the human’s final "approval" is merely a symbolic gesture.

To manage the cognitive burden, military engagement must be defined by three distinct modes:
- Reactive: The system waits for a specific human command (e.g., "analyse this sector"). The human remains the primary architect of the problem.
- Assertive: The system supports a human-initiated task by suggesting categories or drafting options, increasing speed but potentially introducing anchoring bias.
- Proactive: The system initiates alerts and rankings based on its own environmental monitoring. This "pre-frames" the reality for the commander before they are even aware a problem exists, significantly increasing the risk of the human becoming a passive observer.

3. Sovereignty Is No Longer About Land; It is Also About Code

National security is increasingly tied to "computational infrastructure" rather than physical borders. We are seeing the rise of a "Transnational Algorithmic Gap," where nations — particularly in the Global South — risk becoming "computational dependents." These states often import AI systems for defense and policing that they cannot independently audit, containing biases they did not choose.

This isn't a theoretical risk; it is an operational failure mode. A system trained on one continent may fail to "see" a specific landscape or misread human behaviour in another because its training data did not account for local cultural or geographical nuances. When a state depends on foreign cloud providers and opaque code to make lethal decisions, its sovereign authority is partially externalised to the architects of the software.

"In the 21st century, sovereignty may largely depend on control over computational infrastructure... the critical political questions of our time may no longer be who control territory but rather who designs, who owns, and who governs the code." — Barak Oada.

4. The "Starlink Effect": When Private CEOs Hold the Kill-Switch

The borderless nature of AI technology has effectively erased the state’s monopoly on the use of force. The "Starlink moment" — where a private CEO made a strategic military command decision by restricting connectivity in Crimea — highlights a new reality: the emergence of CEO-led strategic veto power. Private corporations now control the critical infrastructure that states depend on for warfighting, allowing commercial actors to dictate the limits of national military operations.

This is driven by "Dual-Use Slippage." Commercial AI tools, such as cloud services from Google or Amazon, can be pivoted into lethal targeting tools almost instantly via API calls. There is no "export event" or physical hardware to monitor; the transition happens invisibly in the software layer. This creates a governance crisis where the infrastructure of modern war is governed by corporate compliance structures rather than the accountability of International Humanitarian Law.

5. The End of the "Pre-Deployment Audit": Security Is a Life Cycle, Not a Gate

The traditional "exam" metaphor—testing a system once before deployment — is obsolete. AI systems suffer from "runtime drift," meaning a tool that is safe on Day 1 may be compromised by Day 30. Security must transition to a model of "Continuous Evidence."
This is necessitated by the "Founder Compromise Paradox." In this scenario, defense "goes negative." A compromised AI tool is not just a "failed" tool; it is an active attack multiplier. It may appear to function perfectly, performing its intended mission while secretly siphoning sensitive data or location coordinates to an attacker. Because these compromises are often invisible, a one-time audit provides a false sense of security.

Meaningful governance requires a standardized, four-stage adversarial lifecycle:
- Threat Modeling: Anticipating adversary behaviour and potential "negative defense" scenarios before the first line of code is written.
- Evaluation: Comparative, evidence-based testing of performance against diverse, non-laboratory environments.
- Monitoring: Continuous, real-time oversight to catch "drift" or the moment a system begins serving an attacker’s interests.
- Remediation: Robust processes for patching, rolling back, or "killing" a system once a vulnerability is identified.

Conclusion: Beyond the "Rubber Stamp"

The overarching reality of 2026 is that "meaningful human control" is at risk of becoming a hollow catchphrase. In systems like Gospel or Lavender, where targeting officers are reportedly processing ten times the number of targets they once did, the human is no longer a "controller." They are "laundering the machine's decision." When the speed of the machine dictates the pace of the human, the human enters the chain only to provide a veneer of legality to a pre-structured outcome.

As researchers like 18-year-old, Bandana Kaur has noted, we are currently gambling with a process that the next generation did not inherit but must live with. If we continue to treat AI security as a gamble rather than a rigorous, continuous discipline, we are not building security; we are building a legacy of unaccountable force.

The question for every policymaker and commander is no longer just whether the AI works. It is this: Are we building a process worth inheriting, or are we simply automating the end of human judgment?
"The discussions are ultimately asking the same question and that is how do we ensure that AI serves humanity rather than undermines it." — UNIDIR 2026.


More On:
- Day 1 of the UNIDIR Global Conference on AI, Security and Ethics 2026.
youtube.com/live/MVD7T0vGw7Y?si=tWC4jeM5fTD6q5vL
- Day 2 of the UNIDIR Global Conference on AI, Security and Ethics 2026.
youtube.com/live/Cf8-MO7DYlE?si=9tnn39BHtnBACwN6
- Vonghour Leng’s Playlist on UNIDIR Global Conference on AI, Security and Ethics 2026.
youtube.com/playlist?list=PLt...

3 weeks ago | [YT] | 0

Vonghour Leng

"We Are Now Entering the Greatest Golden Age of Human's Possibilities" - What Does Sam Altman Mean With His Claim?

In recent years, there was a time Sam Altman — the CEO of OpenAI stated a pressing comment that "we are about to enter, I think like the greatest golden age of human possibility." SO WHAT DOES HE MEAN BY THAT?

Based solely on ‪@vonghourleng‬'s observation, Sam referred to "a massive, AI-driven leap in our civilisation’s capacities." In the ever-changing internet-based globe — this would be translated to "rather than replacing humans, the emerging digital technologies would be drastically immersed and amplified into human's workforce, embarking a new journey to potentially create, problem-solve, and prosper." His ideal vision rests on core fundaments as follows:

1- Unprecedented Problem-Solving
Artificial Intelligence (AI) serves as a tool to tackle humanity's most complex challenges, from discovering new medical cures and solving climate issues to unlocking advancements in physics.

2- Hyper-Personalisation
Each person and/or player will eventually have access to a personalised team of AI experts; which means everyone is able to access to virtual tutors for education, tailored healthcare, and automated coding, and etc.

3- Massive Shared Prosperity
By making intelligence abundant, the cost of creation will decrease, raising global living standards, and allowing widely everyone access to opportunities previously and/or retrospectively only available to a few.

4- Human Skills Shifts
As AI gets more and more capable, competent, and intelligent — they will be able to answer questions instantly, the primary human skill will shift from memorisation to figuring out what questions to ask and setting goals.

Since this had not been explained and/or asserted in any of the author's platforms; despite he sought for an opportunity to write and publish elsewhere in elaboration to his previous videos on "2 Facts You May Not Know About the 21st Century". Therefore, readers are able to catch up on it via accessible attached below:

- Rowan Cheung's Comment: www.linkedin.com/posts/rowancheung_sam-altman-you-…

- Sam Altman's Remark on "How AI Will Impact Our Humanity": https://www.youtube.com/watch?v=4EcSX...

- The Intelligence Age: ia.samaltman.com/

- OpenAI CEO Sam Altman Says This Will Be the No.1 Most Valuable Skill in the Age of AI: www.inc.com/jessica-stillman/openai-ceo-sam-altman…


Catching-Up With ‪@vonghourleng‬ On:
- Medium: medium.com/@vonghourleng
- Substack: substack.com/@vonghourleng
- YouTube: ‪@vonghourleng‬
- TikTok: www.tiktok.com/@vonghourr
- Facebook: www.facebook.com/criticalbyvonghour

#SamAltman #ai #artificialintelligence #goldenage #humanpossibility

1 month ago (edited) | [YT] | 0

Vonghour Leng

Concerning the two Anthropic's AI Models (Claude Mythos and Claude Fable 5) - everyone may be familiar with the very foundational background of the whole story already, I presumed.

Precisely on the other side note, I believe that each individual may have not paid attention on the "Anthropic's Counterarguments" fully well. To break it all down, let us all concentrate on:

(1) Safety and Regulatory Paradox

It goes all the way back to the Official Doctrine to Anthropic's establishment. The corporation was founded on the principle that "frontier AI models pose genuine safety risks and therefore require government oversight and regulation". Their founding CEO, Dario Amodei has publicly stated that:
- Frontier AI capabilities are dangerous if misused
- Responsible AI development requires safety research
- Government has a legitimate role in AI governance
- Companies should work with regulators, not against them.

However, it is quite the opposite when it comes to the actual behaviour. When the US government exercised that oversight by ordering them to either fix the Fable 5 jailbreak or de-deploy the model — Anthropic refused and pushed back publicly. Despite, their given argument was "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people."

-> The Problem

The two positions can not be both true simultaneously. If frontier AI safety is serious enough to require government regulation, then government safety determinations must be taken seriously. You can not advocate for regulation and then reject regulation when it applies to you.

(2) Dispute on Severity

Anthropic's reaction to the government's request ran counter to the company's own public claims regarding the need for AI safety and regulation — a perspective that Dario Amodei again shared this week in a blog post.

Re-considering this through, the government's logic was essentially:
- Fable 5 has guardrails designed to prevent dangerous requests
- Those guardrails were bypassed in 48 hours by researchers
- The system prompt was published publicly
- Therefore, the "safety promise" is broken
- And afterwards, the model should be recalled.

Anthropic stated that "it believes the government became aware of a jailbreak technique that exploits relatively minor vulnerabilities — which other models can also discover such vulnerabilities."

This breaks down to three claims as follows:
1- "The Narrow Jailbreak" It is not a fundamental architectural flaws, but an exploitable edge case. "If all AI models have edge cases, you can't recall every model for having minor flaws."
2- "Other Models Encounter Similar Vulnerabilities" GPT-5.5 and others have similar bypasses. "Singling out Anthropic is an inconsistent enforcement."
3- "People Depend on This Very Model" "Recall" can be interpreted to "having massive disruption for legitimate users." The cure (shutting down access) is worse than the disease (a narrow jailbreak).


WHY THE GOVERNMENT WON THE ARGUMENT

The government's position is internally coherent on the Precautionary Principle Framework (PPF):

1- Anthropic themselves established the principle
By advocating for frontier AI regulation, they endorsed precautionary governance.

2- A cybersecurity exploit ≠ "narrow"
From a national security standpoint, a vulnerability that enables cyberattacks isn't "narrow". It is strategically significant and concrete view to consider.

3- System prompt publication is material
The system prompt is a hidden set of instructions that helps determine an AI model's behaviour. It is unclear exactly how knowledge of Fable's system prompt could be used in practice, but it has drawn attention in security circles. Once this is public, the guardrail architecture is compromised at a fundamental level.

4- Export controls are proportional
The government didn't sue Anthropic or fine them heavily. They only and primarily restricted foreign access to a cybersecurity-capable model. That is a measured response under national security doctrine.


How to Resolve It
To understand the paradox clearly, we need to distinguish between two different governance frameworks:

Framework 1: Precautionary Principle
- Definition: When an activity raises threats of harm, precautionary measures should be taken even if cause-and-effect relationships are not fully established.
- In AI Context: If there is significant uncertainty about safety, restriction deployment is required.
- Applied to the Case: If Fable 5's guardrails can be bypassed, it presents a cybersecurity risk, so it is prominent to restrict it.

Framework 2: Risk Tolerance Approach (What Anthropic practiced)
- Definition: Accept some level of risk if the benefits are substantial and the risk is manageable or limited in scope.
- In AI Context: Not every flaw means and/or requires "model withdrawal". Only egregious, system-level failures justify recall.
- Applied to the Case: A narrow, exploitable jailbreak ≠ reason to shut down a model used beneficially by millions.

Upon careful observation, my perspective towards Anthropic on their corporate behaviour is more likely that "they advocate for regulation, but resist its regulatory application" — as Anthropic's response to the export control is revealing that "they prioritised commercial deployment over the safety framework they publicly champions."

More On:
- Claude Mythos and Safeguards:
www.anthropic.com/claude/mythos
- Project Glasswing:
www.anthropic.com/glasswing
- Project Glasswing: An Initial Update:
www.anthropic.com/research/glasswing-initial-updat…
- Expanding Project Glasswing:
www.anthropic.com/news/expanding-project-glasswing
- Claude Fable 5 and Claude Mythos 5:
www.anthropic.com/news/claude-fable-5-mythos-5
- Statement on the US government directive to suspend access to Fable 5 and Mythos 5:
www.anthropic.com/news/fable-mythos-access
- US Government Orders Anthropic to Pull Claude Fable, Mythos AI Models:
decrypt.co/371027/us-government-orders-anthropic-p…
- Why the US Government Shut Down Anthropic’s Latest Claude AI Model:
theconversation.com/why-the-us-government-shut-dow…

#artificialintelligence #ai #anthropic #mythos #fable5 #aigovernance #cybersecurity

1 month ago | [YT] | 0

Vonghour Leng

As more and more cases of artificial intelligence (AI) usage in the military front appear to happen simultaneously, and impact and risk significantly the lives of thousands and thousands (or perhaps more), it is quite concerning to each individual actor across the region, most particularly to the lives in the virtual world we are currently residing. In the ever-changing internet-based globe, in furtherance, this can be observed and categorised into two types: (1) at the micro-level and (2) at the macro-level - which each actor/party can define and differentiate by themselves accordingly.

However, it is empirically no stagnate to impede the AI booms. Henceforth, it is prerequisite and majorly required to consider and establish a good use of AI with a long-lasting vision practically concentrating on the basics of human rights and humanitarian needs. This being proclaimed, AI usage can be seen through camera surveillance, real-time location tracking, intelligence, and more and more. Owing to which, the potentiality and practical consideration may highly be more on Responsible AI and Ethical AI. BUT WHAT ARE THEY? And How ARE They even DIFFERENT?

Dive Deeper and Learn More From HERE:
1- https://www.youtube.com/watch?v=v1wZw...
2- https://www.youtube.com/watch?v=XP7ah...
3- inss.ndu.edu/Media/News/Article/4454555/controllin…
4- inss.ndu.edu/Research-and-Commentary/View-Publicat…
5- www.aljazeera.com/video/the-stream/2026/3/12/can-a…

To my last remark here, ⁨@vonghourleng⁩ intent to spare some time further navigate and deliver some insights from my reading and/or findings here on TikTok and YouTube. Upon finalising, I will be then gratefully upload and share it here.

STAY TUNED...!!!

1 month ago | [YT] | 0

Vonghour Leng

The very factual concern we are now encountering is, "Is Artificial Intelligence (AI) Replacing Us, Humans?"

Though this is there on the table for the discussion, yet we may all consent that AI is taking our places: doing our jobs, delivering impacts, meeting required outcomes, outsourcing these tools with human strengths and competency, and so on and so forth. However, this may only be just a large proportion. Similarly, the interpretation can be, "AI can not be there all the time and all the places, fixing everything and replacing us." There are things AI can do and can not do, and this is and undeniable and empirical fact.

Therefore and herewith, it is quite dangerous for us, humans. As a matter of fact, we must always remain vigilant and be careful with cautious views to AI. Though they are capable, but they need our command. "SO IT IS STILL IN OUR CONTROL." As of which, our lives are still at risks, even though we could use those tools to its very maximum potential for facilitation. So it is relevant and precautionarily needed for us to be on track to avoid necessary negative circumstances.

In furtherance, it is more contributing taking into account knowing that this is just us and the tools we try and attempt to leverage. Practically as of the video description itself, this is just AI's role as a chatbot and a call-center only. If I have to put this alternatively articulately with the most recent emerging digital technologies, most particularly to AI Twins (Digital Mirrors or Digital Twins) and/or Data's Digital Ghosts - then it could have been a lot more risky requiring us for a more considerate precautions.

With the aforementioned instances, that goes without saying that AI, though as a tool could be endangering and violating its very terms of invention impacting us directly and indirectly. To its core, the fundamental issues remain to be affecting our data and privacy. Image having digital mirrors doing the content for us, yet "WHAT IF" it malfunctions and generates inappropriate contents instead - how would the other stories go? "AND THIS IS JUST A SIMPLE WHAT IF SCENIO, and WHAT IF THERE ARE MORE OF IT TO CONSIDER?"

#whatif #artificialintelligence #AI #jurisprudentialresponses #precautions #whatifscenio

2 months ago | [YT] | 0

Vonghour Leng

🔐 The AI they built — and then locked away.

Anthropic just released Claude Opus 4.7 to the public. But the model everyone is actually talking about is the one they didn't release.


It's called Claude Mythos. And here's why it matters:
✔️ It autonomously discovered thousands of zero-day vulnerabilities in every major operating system and web browser — including flaws that survived 30 years of human review.
✔️ When tested by the UK's AI Security Institute, it succeeded on expert-level cybersecurity challenges 73% of the time — tasks no AI could even attempt a year ago.
✔️ It triggered an emergency U.S. government briefing — the Treasury Secretary and Federal Reserve Chair called the CEOs of every major American bank to warn them.
Anthropic's response? Keep Mythos restricted to 40 partner organizations, and release Opus 4.7 instead — a model with its cyber capabilities deliberately reduced and automatic safeguards built in to block high-risk uses.


Opus 4.7 is still the most capable publicly available AI for coding — outperforming GPT-5.4 and Gemini 3.1 Pro on key benchmarks. But it's built to be governed.
As an AI governance researcher watching this from Southeast Asia: the Mythos case is a preview of the choices every government and institution will face very soon. The AI capability frontier has outpaced the governance infrastructure designed to manage it.


The question for ASEAN, for Cambodia, for the July 2026 Global Dialogue on AI Governance in Geneva — is whether that governance can be built fast enough.


📖 Full analysis on Substack and Medium (link in bio).


Sources:
- Anthropic, Project Glasswing · anthropic.com/glasswing · April 7, 2026
- Anthropic, Introducing Claude Opus 4.7 · anthropic.com · April 16, 2026
- CNBC, Anthropic rolls out Claude Opus 4.7, less risky than Mythos · April 16, 2026
- TechCrunch, Anthropic debuts preview of powerful new AI model Mythos · April 7, 2026
- UK AI Security Institute, Our evaluation of Claude Mythos Preview's cyber capabilities · aisi.gov.uk · April 14, 2026
- Axios, Anthropic releases Claude Opus 4.7, concedes it trails unreleased Mythos · April 16, 2026
- Fortune, Anthropic says it is testing Mythos after data leak · March 26, 2026
- Bloomberg, How Anthropic discovered Mythos AI was too dangerous for release · April 16, 2026
- Benzinga, Treasury Secretary Bessent and Federal Reserve Chair Powell brief bank CEOs on Mythos · April 16, 2026


#AIGovernance #ClaudeMythos #Anthropic #Cybersecurity #DigitalPolicy #ASEAN #ArtificialIntelligence

3 months ago | [YT] | 0

Vonghour Leng

***Open Opportunity for ASEAN Students, Professionals, and Early Career!

YSEALI CyberSafe ASEAN Call for Applications! 🌟


🌐 Are you ready to tackle the growing threat of cybercrime in the region?
Now is your chance to join the YSEALI CyberSafe Regional Workshop and be part of the solution!


📅 What’s in it for you?
(1) Foremostly, you get the greatest chance to learn best practices from U.S. experts. (2) You time to work collaboratively with peers across ASEAN.
(3) Last and not least, you may have your spotlight to engage in regional cooperation in combating online scams.
***Together, we can strengthen cybersecurity, protect victims, and build a safer digital future!!!


Apply now and make a difference! 🛡

🔗 yseali-cybersafeasean.org
📅 Applications Deadline: December 12, 2025, at 23:59 GMT+7

© YSEALI


#YSEALICyberSafe #ASEAN

Skip the Carousel

1 2

8 months ago | [YT] | 1