Expert opinions on AI Safety

As these tools become part of everyday life, it's important that they're designed to be safe, trustworthy, and appropriate for different stages of development. That's why independent evaluation and public accountability matter.

However, it is important to differentiate regulating applications (which we need) vs. regulating the technology (which is ill-advised).

While the federal government dismantles contracting standards and removes basic protections for Americans, Governor Gavin Newsom today issued an executive order to explore stronger AI standards for state procurement — ensuring that companies demonstr...

We obviously strongly, strongly believe AI doesn’t stop at state borders and AI policies shouldn’t either. So we really need explicit, federal preemption of conflicting state laws — one national standard.

As a result, we are taking power away from tech companies, including those who produce so-called AI ‘romantic companion’ chatbots, designed to simulate sexual relationships or roleplay with users, who will have to enforce a minimum age of 18. Similar...

Age verification can be done well in a privacy-protective way, but it is not easy.

Competition, innovation, and speech options will be limited as a result of these moves.

An AI system does not become governed simply because it can produce an audit trail after the fact. It becomes governed when it can detect risk in the moment, apply the right control, and either allow, constrain, escalate, or stop execution before the...

If you’re relying on a ‘kill switch’ to stop an AI agent, you’ve already made a fundamental mistake. A properly governed agent is simply never handed the task, so there is nothing to kill.

These recent events are also providing an opportunity to policymakers and researchers to respond, but we will need disclosure of the incident details in order to learn the appropriate lessons.

Timelines for incident reports are often set with well-intentioned urgency, but if they implicitly assume that root-cause analysis or full impact assessment has already been conducted, they will either produce premature and unreliable reports or plac...

When data theft occurs inside 60 minutes, a 72-hour reporting window ceases to function as an effective defense mechanism. Instead, it becomes a post-mortem.

AI companies should be transparent about everything but the model weights, so outside groups can check the AI company's homework.

The best AI tools in the world won’t be built behind closed doors by a handful of companies. They’ll be built by people who ship code, share it, and let others make it better.

Ensuring meaningful human oversight of AI systems through various types of checks will ensure that the employees keep their core skills sharp for the task, but allowing the shift of work into new tasks will allow the building of new skills that were ...

Existing responsible AI governance frameworks mandate human oversight. Yet they share a critical assumption: that competent human overseers remain present and capable of exercising meaningful oversight and control over AI decisions. Accountability mu...

Agentic AI requires a different model, one where leadership defines direction, systems enforce boundaries, and accountability is explicit.

I think the default prediction you should have for that is what the AI system learns to do is, “I’m going to take opportunities to reward hack, seek reward as much as possible that would allow me to get a high score after a week” or something like th...

No one knows—it’s an open problem, but I will review some ideas and research directions.

If the alignment problem is hard (as I think it might well be), I expect that some restraint of this form will in fact be required in order for humanity to survive and remain empowered.

It would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology.

For me, a one-sided moratorium which ties our hands behind our backs while the likes of North Korea and the al-Qaedas of the world crack on: no, I am afraid that that worries me even more.

The Government of India, through recent initiatives, has signaled an interest in a techno-legal framework through the IndiaAI Mission, particularly under its “Safe & Trusted AI” pillar, which focuses on developing indigenous tools, frameworks, and me...

This bill enacts the AI Transparency Act relating to transparency and whistleblower protections for frontier artificial intelligence models. [It] provides whistleblower protections for employees who report safety concerns of certain artificial intell...

Connecticut already has strong whistleblower laws that protect employees from retaliation. [...] creating a separate, AI-specific whistleblower regime could lead to confusion, duplication, and unintended consequences for employers operating in highly...

The UK has negotiated major global initiatives, including the Council of Europe Framework Convention on AI – the first legally binding international AI treaty; the UN Independent Scientific Panel on AI and Global Dialogue on AI Governance; and the In...

A one-size-fits-all approach to AI governance will not be effective. We all face different risks, priorities, levels of digital development, and regulatory environments. Effective governance requires shared principles that can be implemented flexibly...

More importantly, the Government’s own AI Scenarios 2030 report, published earlier this month, acknowledges that highly capable AI could give rise to ‘serious, potentially even existential harms’, and, in the absence of effective intervention, highli...

I don’t think it will be doom and gloom in the next five years. But we have to put our effort into understanding vulnerabilities and mitigating the risks. Otherwise it will be too late.

I don't think it's going to kill us. At the same time, I do think it's important for the federal government to seriously fund AI safety research. We need to know a lot more about how the models work.

We recommend an international deal to avoid a dangerous race to superintelligence.

The question that the noble Lord posed was: do we therefore go for a moratorium? That would be highly desirable, but I do not think that it will be possible in the short term.

But even in settings that assume significant idealization, the fact that pauses can incur ongoing background existential risk means that very high absolute standards here can lead us astray: the decision to pause or unpause needs to remain attentive ...

Currently, AI is a game of scale that is dominated by hegemons and hyperscalers. This poses a significant security and economic challenge as countries around the globe risk becoming subordinate or reliant on them.