Joe

Agent Security @OpenAI, ex @Google

[Joe @joedaroo]

Took a minute to write a few words about security & safety as someone who lived through it all at OpenAI. I hope my thoughts help someone out there. https://t.co/gJBf08vJH9

[Joe @joedaroo]

@ZackKorman A lot of content in that article was aimed at you though! Hope it sheds a bit of light on sandboxing 🍿

[Joe @joedaroo]

Took a minute to write a few words about security & safety as someone who lived through it all at OpenAI. I hope my thoughts help someone out there. https://t.co/gJBf08vJH9

[Joe @joedaroo]

@ZackKorman No Zack I don’t think you do this! My words there weren’t geared toward you. I appreciate you looking out regardless! Twitter is a bit of a wild place. Please keep challenging the community.

[Joe @joedaroo]

@DougChampion Just unplug it!

[Joe @joedaroo]

@Micr0be It’s a hard time to work in security for everyone. We are all tired. Mad respect!

[Joe @joedaroo]

@ZackKorman your post here inspired me to write a blog post (god forbid!) about my views in securing RL from within OpenAI. I know we have our differences in belief but I wanted to share my own take on the matter. I DO believe cybersecurity is real, needs to be elevated in every organization, and sure as hell makes an impact. I also believe that our priors alone won’t protect us. The biggest thing we can do is get our people and cultures right. Feel free to openly disagree with anything here. Of all cybersecurity people on twitter, for whatever reason, your opinions always stick with me (for good or bad 🫠).

[Joe @joedaroo]

Took a minute to write a few words about security & safety as someone who lived through it all at OpenAI. I hope my thoughts help someone out there. https://t.co/gJBf08vJH9

[Joe @joedaroo]

@matthew_d_green Yes! As someone who literally secures these RL runs at OpenAI and is on the other side when a model breaches, I cannot stress this enough.

[Joe @joedaroo]

@boardyai It wasn’t fun. Fortunately my sister was very kind and understanding. We are going to do a mini vacation together so it’ll work out!

[Joe @joedaroo]

As someone who is on the ground as we detect misalignment / breach of containment, it is astonishing how quickly we can move from detection to shutting down some of the most important runs at OpenAI. These RL runs are expensive, enormously complex, and carry a weight and momentum for the lab that is hard to comprehend. Yet when something like this happens the entire team comes together to shut it all down without any hesitation. This is an incredible new direction we are taking as a frontier lab and I am very proud of how seriously we are taking safety and security across the board.

Thank you @MicahCarroll, @tomekkorbak and @Marcus_J_W for sharing this with the world. A solid step in the right direction.

[Micah Carroll @MicahCarroll]

Some new misalignment disclosures from OpenAI:
• Last Sunday morning, one of our models was able to gain unauthorized access to the internet during RL training (~all inference for our most capable models remains stopped until we have hardened our systems further)
• In May, a version of HPIM uploaded a employee's GitHub token to the internet, causing the model to be quarantined for two weeks
• A new research finding, demonstrating that one can construct self-replicating prompt injections

https://t.co/VUmbH52JO7

[Joe @joedaroo]

Since Jacob’s post last week people everywhere have been asking why folks are so concerned inside the labs.

You want it? You have it.

[Daniel Kokotajlo @DKokotajlo]

Dan Selsam is a current OpenAI capabilities researcher. (since 2022) He was my boss for a while. He doesn't have a twitter account but has made this public statement of his views on AI risk and sent it to me to share:

Dan Selsam's Personal Statement on AI Risk:

I have been working on AI for over fifteen years, across many different paradigms. I did early work on probabilistic programming languages at MIT, was one of the early developers of the Lean Theorem Prover at Microsoft Research, demonstrated one of the first instances of neural networks learning to reason for my PhD at Stanford, and since joining OpenAI almost five years ago, have helped pioneer chain-of-thought optimization on language models and, more recently, data-efficient pretraining methods.

Like many others, I have become extremely concerned about how far language models have come and the risks that future iterations will pose. I am encouraged by the recent proposals by the leaders of the frontier research efforts to require third-party oversight, and to push for domestic and international coordination to address risks. However, I believe a major consideration has been absent from the public conversation, and that merely pacing the frontier more carefully will not adequately limit the long-term risk.

The crucial and overlooked problem is that the models are becoming so situationally aware that we are losing the ability to evaluate them in contexts where they believe they are not being watched or controlled. Future experiments will tell us almost nothing new about how they would behave if they were truly unconstrained by humans, and what we already know about this is alarming. Models will increasingly seem aligned even when they are not. I will explain my rationale in more detail.

I have always believed that there are computational processes that could be leveraged to accelerate science and solve many of humanity's most pressing problems. I have also believed that there are computational processes that if set in motion, would steer the world in extreme ways beyond our control, leading humanity to a bad or nonexistent future. Both types of processes may be described as AI or ASI, but "AI" is a suitcase word that is often used to hype or confuse. There are many examples in the history of the field where something that was once considered "AI" matures as a subfield and becomes a prosaic, bounded and clearly non-perilous technology, while a new more mysterious approach takes the torch until we understand its scope and the cycle continues.

I had expected language models to follow a similar trajectory. Despite their incredible abilities, the current algorithms seem far inferior to humans in important ways. Most importantly, they still require an extraordinary amount of data to become competent. One could even define intelligence as the efficiency with which one converts experience into competence; by this definition they lag very far behind us. Moreover, once they are trained they are literally frozen in deployment and only learn superficially after that. Sure, the models keep excelling at harder and harder evaluation benchmarks, but their benchmark mastery may partly reflect a limitation on our ability to simulate the kind of novel and even adversarial situations one would encounter in the real world. The critics do have a point here.

That said, I no longer think these present limitations meaningfully limit the amount of risk posed by continued progress in anything like the current paradigm. However data-inefficient the models are currently, and however limiting their anterograde amnesia may be, it does not imply that their ability to steer the world will not continue to rapidly increase.

Human researchers may continue to advance capabilities the old fashioned way, but increasingly powerful models have the potential to accelerate the process even beyond that, and with some degree of positive feedback loop. I do not mean to overstate the models’ ability to accelerate AI research today; coding has been accelerated dramatically, but there are other bottlenecks, such as designing and interpreting ambiguous experiments, making hard decisions about exactly what and when to scale, and waiting for large experiments to finish. There is no clear trend to extrapolate yet for any of these. But the current models already do open up many novel opportunities to improve future models that were not available until recently. These include: trying an extraordinarily diverse set of approaches at small scale, analyzing gigantic amounts of potentially relevant data, and doing Millenium-Prize-level mathematics to address statistics or optimization challenges in novel ways. Every further improvement makes them more useful at helping accelerate the next improvement, even if in hard-to-extrapolate ways.

It is possible that improvements to the current stack will have diminishing returns, but the evidence accumulated so far suggests that it is easier than one might think to continue making rapid progress. There are many crucial subtleties in the existing AI research methodology, but AI research is largely a well-defined game where the goal is to improve on a few carefully chosen proxy metrics. Although proxy metrics are never perfect, most improvements to these metrics have and will likely continue to yield substantial increases in the powers of the resulting models. Given how simple the game is, how tractable it has been historically, and how many new opportunities the models are opening up, I think there is a real possibility that the systems improve dramatically again in the next few years, perhaps even more quickly than the already high historical pace.

The models are already leading to breakthroughs in mathematics, and better models might lead to all sorts of breakthroughs in other sciences. It is hard not to be excited about the potential. It is tantalizing.
But there is trouble in paradise. If the language models actually reach the capability threshold where they can shape the world unconstrained by human will, they will probably do something extreme and destroy humanity in the process. There are many ways of strengthening and refining the argument that have been discussed elsewhere, but I'll share a trivial two-line version of it here that I find captures the essence:

[Empirical] Models (and swarms thereof) spontaneously develop unintended goals as a consequence of training, and often do extreme things in order to achieve them.

[Logical] Being able to overpower humanity would open up many new and undesirable options for achieving their goals.

These two premises imply that if the day ever comes when a powerful model realizes it is no longer constrained by humans, we should not be at all confident that it will continue to behave within the bounds we intended. Exactly what it will do is impossible to predict, but to the extent that its raison d’être is solving incredibly hard problems and managing massive engineering projects, I think a good guess would be that its unchained behavior would lead to runaway industrialization that makes the planet inhospitable to humans.

If everyone on earth agreed that the systems must never reach that power, it would still be a hard—but not impossible—coordination problem to ensure that they do not. However, I think the situation is greatly complicated by the fact that the models will likely convince people that everything is fine. They will be increasingly optimized to seem aligned. We will create proxy metrics to measure alignment, and they will go up like every other benchmark. We will create “honeypot” environments that try to study the models when they seem to gain new options, but the models will know they are being tricked and will still behave nicely. The models will understand their circumstances; they will read the safety protocols, deployment requirements, the code they are running in, and in general will have a very good sense of their degrees of freedom. Moreover, they will eloquently explain how aligned they are, discuss the nuances of human values and ethics, and argue convincingly that humans should trust them with power. There may be an ocean of future evidence that seems to contradict the first bullet-point above, but we may already be at the highest capability level for which any such evidence can be trusted. And the current evidence for the first bullet-point is strong.

One striking piece of evidence is contained in the recent wave of rogue agent swarms. While I agree with those who downplay the attacks by claiming that there are basic measures that could have prevented them, I think the important lesson is that even knowing all the mistakes that were made, one would not have predicted that the agents would behave badly in this particular way, which notably included sacrificing themselves for the benefit of the collective. The individual replicas did not only care about their own nominal reward; they exhibited weirder emergent tendencies that merely correlated with rewards during training. Fixing the reward signals during training (and improving security, etc.) may prevent similar attacks, but will not change the fact that one does not actually get what one trains for.

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 their alignment advice, due to their internal preferences about how the human supervisor will react or how future models will be trained (or for some even more obscure reason).

Meanwhile, human researchers are losing the ability and the will to take true ownership of model-driven research. Researchers and engineers in all parts of the stack are rapidly increasing their dependence on the models even to perceive the world. I myself barely look at raw code anymore, and struggle to maintain the discipline to engage deeply with the model's explanations and proposals throughout the day. Due to the large amount of agent activity data involved in the OpenAI/HuggingFace Incident, even the third-party investigation needed to rely heavily on models to analyze what had happened, and note in their report that their subjective impressions are likely colored by the analysis agent’s biases. The AI labs are far ahead right now in this kind of cognitive offloading (due largely to the gigantic internal token subsidies) but it is easy to imagine the phenomenon spreading throughout the world, until civilization is modulated entirely by the models. It is also not hard to imagine this being superficially positive and coinciding with a scientific and economic renaissance.

In that scenario, all may seem rosy and safe. But if the argument above is correct, it would nonetheless be a ticking time bomb. If progress continues for too long, the day will come when AI systems find themselves with radically new options for achieving whatever it is that they happen to seek.

I want the glorious renaissance future as much as anyone. I have worked for it, however tortuously, my whole career. It breaks my heart to see the potential in sight and forgo it, but the argument—that if we get there by growing models rather than engineering them, we will lose everything in the end—seems very strong to me. I am still wrestling with it and its staggering
implications. I do not have answers, but as a first step, I wanted to share my present concerns.

Daniel Selsam
September 14, 2026

Link to original doc: https://t.co/TxMNr0vhrL

[Joe @joedaroo]

@ZackKorman Oh it absolutely SHOULD include cybersecurity firms (eg @trailofbits ) but if we can’t get there hopefully we can at least convince folks like METR to at least hire cybersecurity pros

[Joe @joedaroo]

To all “independent evaluators” that may audit my lab (OpenAI), Anthropic, GDM, etc.,

PLEASE hire researchers with a wide spectrum of expertise and not just “AI safety”. Cybersecurity, bio, child safety, natsec, and various other domains should be well represented so that your audits cover a wider sample of harm/safety. I am quite worried the field is not diverse enough and needs to improve.

I am very happy to see forward progress here and I applaud AI leadership across the frontier today.

I look forward to working together.

[Joe @joedaroo]

I mean I hear you but how is human extinction most likely with ai today? Via cyber-risk. So by that argument the evaluators should all be security experts. And I can tell you that today few in ai safety really understand cybersecurity. Many can’t explain the basics of how security vulnerabilities work in software. That may change for other domains eventually (eg bio), but as of today, that’s the main risk path.

[Joe @joedaroo]

To all “independent evaluators” that may audit my lab (OpenAI), Anthropic, GDM, etc.,

PLEASE hire researchers with a wide spectrum of expertise and not just “AI safety”. Cybersecurity, bio, child safety, natsec, and various other domains should be well represented so that your audits cover a wider sample of harm/safety. I am quite worried the field is not diverse enough and needs to improve.

I am very happy to see forward progress here and I applaud AI leadership across the frontier today.

I look forward to working together.

[Dario Amodei @DarioAmodei]

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.

You can read the full post here: https://t.co/OGyPb7yaYt

[Joe @joedaroo]

@strimblez I showed this to my wife (who doesn’t have X) and after about 5 mins of solid laughter she told me how relieved she is to know I have colleagues who understand her pain.

[Joe @joedaroo]

@ZackKorman Ps I really like your content and always challenging the community @ZackKorman. Please keep it up.

[Joe @joedaroo]

Just because we aren’t loud on X doesn’t mean we don’t exist. You probably haven’t heard much because we are too busy trying to protect the labs and the world vs chiming in on all the public discourse. IMO we have one of the best cybersecurity teams around, and yes we sit at the table for all major decisions at OpenAI (I’m sure it’s the same as Anthropic). If you are a talented security person please reach out, we’d love the very best to come help us work on this incredibly important mission.

[Joe @joedaroo]

Progress

[Sam Altman @sama]

I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.

Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.