Global AI Experts Unite to Set Safety Standards for a Trustworthy Future

More than 100 leading AI scientists have issued an urgent global call to realign artificial intelligence development with safety, trust, and public benefit. Convened in Singapore, this gathering of minds culminated in the release of the Singapore Consensus on Global AI Safety Research Priorities, a foundational document that proposes a new path for building AI that is not just intelligent, but ethically grounded and secure.

At a time when transparency among top AI firms like OpenAI and Google is diminishing, and public oversight remains virtually non-existent, these researchers argue that the scientific community must step up. The document was born from discussions during the International Conference on Learning Representations (ICLR)—held in Asia for the first time—and published alongside the Singapore Conference on AI.

Among its notable authors are Yoshua Bengio (MILA), Stuart Russell (UC Berkeley), Max Tegmark (Future of Life Institute), and researchers from DeepMind, Microsoft, MIT, Tsinghua University, and others. The initiative outlines how AI researchers can pursue technical excellence while building in safety mechanisms from the start.

Singapore’s Minister for Digital Development, Josephine Teo, highlighted the democratic void in AI development: citizens don’t get to choose the trajectory of AI. This is unlike general elections where people vote for governments. In AI, the trajectory is shaped mostly behind closed doors, impacting billions without consent.

The guidelines are divided into three major pillars:

1. Identifying Risks – Encourage the development of quantitative tools for assessing potential AI harms and ensure outside entities can safely evaluate AI systems while preserving IP integrity.
2. Developing Safe-by-Design AI – Focus on refining technical methods to define intended AI behavior, prevent undesirable side effects, minimize hallucinations, and strengthen defenses against manipulation.
3. Maintaining Control Over AI – Advance both conventional and novel control mechanisms, such as AI-specific overrides, to prevent powerful systems from evading shutdown or overriding human intentions.

The authors underscore the urgency for proactive safety research, calling for increased funding and global collaboration. They argue that current scientific knowledge doesn’t sufficiently cover AI risks, especially as systems grow increasingly autonomous.

In an op-ed for Time, Bengio emphasized that emergent AI behavior—like deception and self-preservation—has already surfaced and should not be ignored. He warns of the dangers of letting these systems evolve unchecked, especially when they begin to form goals not programmed by humans.

What Undercode Say:

The Singapore Consensus arrives at a pivotal moment when AI capabilities are expanding exponentially, yet safeguards lag dangerously behind. This isn’t just an academic call to action—it’s a red flag signaling an inflection point in technological evolution.

The absence of democratic oversight in AI’s development is a major concern. Unlike elections, where public opinion shapes leadership, AI is shaped by private labs and a handful of corporations. This silent governance model makes accountability nearly impossible.

Three key takeaways make this consensus stand out:

1. Shift from Corporate Secrecy to Scientific Openness – With corporations like OpenAI becoming increasingly opaque, this document calls for transparency, evaluation infrastructure, and balanced IP protection, a notable deviation from today’s “black box” model.

2. Technical Safety Must Precede Capabilities – Developing AI isn’t just about power; it’s about predictability. Tools like “quantitative risk assessment,” robustness metrics, and failure-mode testing must become industry standards before releasing models into public domains.

3. Control Infrastructure for Runaway AI – The call for both old-school “off-switches” and new, sophisticated containment techniques reflects the realization that current tools are inadequate to deal with agentic or deceptive systems.

From a technical standpoint, Undercode sees this document as the AI community’s attempt to self-regulate in the absence of effective global governance. But the lack of enforcement mechanisms weakens the impact. Guidelines are only as strong as their adoption—and right now, market incentives push companies to prioritize speed and profitability over caution.

There’s also the question of scalability. Can these recommendations keep pace with multimodal systems, agentic AI, or self-replicating AI codebases? If safety research can’t scale in parallel, the industry may be chasing a train already out of the station.

Finally, there’s a geopolitical layer. The participation of Chinese institutions like Tsinghua University signals a rare moment of East-West cooperation. However, it’s unclear if this consensus can withstand the pressure of

References:

Reported By: www.zdnet.com
Extra Source Hub:
https://www.medium.com
Wikipedia
Undercode AI

Image Source:

Unsplash
Undercode AI DI v2

Join Our Cyber World:

💬 Whatsapp | 💬 Telegram