AI alignment is solvable

I believe it’s an engineering problem ... and we all need to hope that’s an engineering problem. If it’s not an engineering problem, it’s not solvable.

Current and near-future AI systems are tools, alignable through engineering.

Alignment is not solved but it increasingly looks solvable [...] Since then a lot has happened: pretraining continued improving and RL became a much bigger deal.

We treat misalignment as a central, but solvable, problem for both us and for the field.

This committee has discussed ideas such as third-party testing, a new national agency, and an international coordinating body, all of which I support. Here are some more ways to “move fast and fix things”: Eventually, we will develop forms of AI that are provably safe and benefi…

I believe that we can and must solve technical alignment, and that interpretability is the way there.

If these ingredients work as hoped, the resulting technology would in principle let us describe the algorithms inside a model as it is trained, flag deceptive alignment and reward hacking, and train against those flags to produce an aligned system while paying a manageable alignm…

Understanding scheming is what makes alignment solvable.

How computational agent theory leads to a solution of the technical alignment problem, something that requires a major rethinking of how we design AI systems.

I'm now very confident that it is possible to build AI systems that don't have hidden goals, hidden agendas. [...] A Scientist AI would be trained to give truthful answers based on transparent, probabilistic reasoning.

Objective-driven AI systems will be safe and will remain under our control, because we set their objectives and guardrails and they can't deviate from them. Agentic systems can plan action sequences to accomplish a task, subject to safety guardrails built into their architecture.

Current LLMs are intrinsically unsafe, but with world models, guardrail objectives can be implemented — so by construction, they will not knowingly produce actions that will produce dangerous results.

Overall, our impression is, as we hypothesized in our discussion of Claude's constitution, that teaching the principles underlying aligned behavior can be more effective than training on demonstrations of aligned behavior alone.

Negative constraints ("what is wrong") encode discrete, finite, independently verifiable prohibitions that can converge to a stable boundary.

The correct target is not a self-preserving system under external constraint, but a system constitutively indifferent to its own continuation -- Existential Indifference (EI).

MAP-AI, a new control-plane system architecture for aligned decision-making under uncertainty, enforces alignment through Monte Carlo estimation of outcome distributions and admissibility-controlled policy selection rather than static model-level constraints.

It can and will be mitigated if the right research is done and proper precautions are taken and we are not racing at absurd speeds — better models will help solve alignment.

Alignment by default (in any of the above senses) does not hold, but alignment is tractable enough that some combination of targeted technical research, non-default but plausible policy interventions, and elevated (though not unrealistic) safety concerns and altruism among AI lab…

The alignment problem is solvable in principle. Whether it is solved in practice depends on whether the necessary resources, talent, and institutional commitment are mobilized before the systems that need to be aligned become too capable to correct.

As AI systems become more capable and more autonomous, alignment has to keep pace. The hardest problems won’t be solved by any one organisation working in isolation—we need independent teams testing different assumptions and approaches.

So I think we'll solve it if we get our act together, we do this internationally, put all the best minds on it, we get going now... And I think given sufficient time with sufficient brain power — I believe in human ingenuity — I think we'll get this right.

Fundamentally, we just do not know how to do alignment. If somebody came with a thousand-page document that explained exactly how to do everything, we couldn’t get a model to learn exactly that.

We don't believe the AI industry has solved alignment and monitoring to a sufficient degree to responsibly scale at maximum speed.

None of these approaches that you'll often hear mentioned—interpretability, testing or alignment—is technically adequate. We don't know how to build superintelligent AI safely. The plan is basically to roll the dice.

We don't know how AIs are aligned. A somewhat crazy aspect of the current situation is that we have very little confirmed public information about why frontier AIs end up being apparently behaviorally aligned. And more generally, we don't know what factors in training are most re…

The honest position is that we do not know if alignment is possible, we do not have a path to solving it, and development is proceeding regardless.

All of these approaches are terrible. No one has a plan, no one is making any meaningful progress towards anything that even resembles alignment. [...] Alignment is way too hard, maybe impossible. [...] The product work was an attempt to make Conjecture make sense after we stoppe…

AI alignment is impossible. Not difficult — impossible.

My argument is that it's impossible to do that. You cannot indefinitely control something smarter than you. So it's not a question of more money or more time or any other resource.

We prove that no verification procedure can simultaneously satisfy three properties: soundness (no misaligned system is certified), generality (verification holds over the full input domain), and tractability (verification runs in polynomial time). Each pair of properties is achi…

What this proof shows is that there is no finite set of guardrails that is universally robust against adversarial prompts.

While we have shown that sufficiently strong AI cannot be fully controlled or predicted, we also demonstrate that agents can be influenced by other agents without central control, and that greater diversity and openness influence their behaviour.

Many believe it’s impossible. Just like it’s impossible for chimpanzees to control us. So most likely if we build superintelligence, its the end of the era where humans are in charge of Earth.

The conventional view is that these concerns are about technology, and that if we take enough time then we can perfect or align the technology to produce a single benevolent superintelligence. I think this view of alignment is fundamentally flawed. Humanity is not a monoculture. …

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

AI alignment is often framed as the task of ensuring that an AI system follows a set of stated principles or human preferences, but general principles rarely determine their own application in concrete cases.

AI alignment is not something that works in theory but is difficult to put into practice. It's something that doesn't work in theory.

If we can't align on these questions among humans, how do we encode them for AI systems that might work on recommendations to optimize across several human civilizations?

Four of six dimensions show structural failures. Two of the four appear tractable to institutional design; the other two, the public reason problem under cognitive incomprehensibility and the non-domination problem under permanent capability asymmetry, call for new normative theo…

A more fundamental fix would be to train the models not to cheat in the first place – but given this kind of behaviour was reported in frontier models more than a year ago, robustly aligning it away may not be easy.

These results suggest that as models scale, their interpretation of prompts becomes more sophisticated but also less predictable, posing a significant challenge for robustly steering model behavior and highlighting the need for compositional and model-specific alignment technique…

Jailbreaking is a clear example where current systems do not converge on human values, even when those values are simple, abundant in training data, and extensively reinforced. This is real evidence against the claim that alignment is easy or natural.

The panel cautions that there is no scientific guarantee that autonomous agents will follow their instructions and that evidence of them departing is already accumulating. Accountability and oversight must therefore be built by design and not reconstructed from failure.

The problem of mimicking human misbehavior is probably unsolvable with current LLM design and training.

Alignment cannot be “solved” through technical design alone, but must be managed through ongoing institutional processes that determine how objectives are set, how systems are evaluated, and how affected communities can contest or reshape those decisions.

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.

The dominant question—the organizing concern of most safety work—is alignment: how do we get AI systems to pursue the right goals, embody the right values, behave within acceptable limits? This question is treated as foundational. Get alignment right and the hard part is solved. …

The alignment problem is not primarily a values problem. It is an existence problem.

I think that could work. I don’t think we have a strong argument that the pragmatic mixture of approaches will get all the way there, but it just seems very dicey, and our understanding of the dynamics involved is very weak.

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

If it takes over that's what will happen. [...] We don't know how to do that. We don't even know if it's possible.

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.

Alignment is supposed to be hard too, and it may well be. But the usual arguments assume we can’t see inside models. We assume we get one training setup, and it has to be right in advance. If we can read and understand the system, maybe we can make simple, careful changes instead…

Many AI researchers grant these concerns and recognize that the hard version of the alignment problem is unsolved; however, they generally believe that the better models of the future will help solve it. I fear we may already be near the point where models systematically bias the…

Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.

First, we need some degree of coordination among the leading development efforts to ensure that the development of superintelligence occurs in a manner that allows us to both maintain safety and help smooth integration of these systems with society. There are many ways this could…

The safety standards are set unilaterally by each company, with limited external accountability. And the entire framework assumes a Western liberal democratic value system that is not universal.