Expert opinions on AI Risk

Future risk reports, system cards and external evaluations should be assessed for reproducible methods, clear separation between simulated and real actions, disclosure of missed detections, and evidence that safety interventions do not simply make pr...

For the most advanced frontier AI models, an independent, federally overseen, industry-backed organization could set safety standards and verify independent, voluntary audits, similar to how other critical sectors are managed.

Alignment cannot depend on one AI becoming perfectly trustworthy.

The relational alignment approach proposed here has not been subjected to adversarial testing comparable to the studies it cites. Until that testing is conducted, the proposal remains a theoretical alternative, not a validated solution.

The core claim of this contribution is that strong ethical alignment is conceptually misguided and that weak alignment will remain limited.

No safety system can eliminate every dual-use risk. Once model weights are public, no developer can guarantee control over every downstream modification or use. Model-level safeguards can raise the barrier to abuse, but they cannot provide absolute c...

Restricting access to a dual-use AI model is precautionary only if it delays harmful actors more than defenders. That condition varies across actors: a state agency or organized criminal group may obtain a substitute through theft, distillation, inte...

Shared access to open-source AI systems is important for protection in a world where AI can be used to hack into electronic systems, like voting systems and benefits infrastructure.

If mandated AI safety reviews are to have any teeth, they must be accompanied by a ban on dangerous AI models that have failed their assessments. This means that some method must be found to restrict access to dangerous open weight AI models. [...] C...

A safety regime that restricts legitimate defenders, while capable models remain available for attackers, creates an asymmetric disadvantage.

Every action and every outcome needs to be clarified and justified to a board, an auditor, or a customer.

We have a very natural instinct to say: we want as much interpretability as possible. One of the sci-fi worry cases is that agents start speaking in languages to each other that we don't understand. [...] I bet there are ways to define boundaries tha...

Explainable AI (XAI) is often offered as the answer to the black box problem and is broadly defined as “machine learning techniques that make it possible for human users to understand, appropriately trust, and effectively manage AI.” Around the world...

There is a significant chance of conscious LLMs within a decade.

We will not just one day have an easy answer where everybody will agree.

A conscious being isn’t a word-guessing app that knows more words and has more computing power to guess with.

There are a lot of potentially good outcomes from AI. I think the most likely outcomes are good. But the risk of bad and even catastrophic outcomes is high enough to be worth taking seriously. We should actively build safety mechanisms in case we nee...

Plot the same improvements on a linear y-axis and you get steady incremental gains, impressive but not the hockey stick that triggered the panic. [...] We don't have evidence that Mythos is actually an important step towards broad superintelligence.

I would like to see a CERN for AI [...] where all the top scientists come together and see whether they can make a super intelligence. And, if they can, they contain it into a system where it can't just go out and persuade people to let it run the wo...

In the face of these threats, I urge the Government to take the following steps: first, to publicly recognise that superintelligence poses an extinction threat to humanity; secondly, for the UK to prevent the development of superintelligence on its s...