Bartosz Brachaczek

[Jon Oberheide @jonoberheide]

Why are people acting like it's @OpenAI/@AnthropicAI's fault that vulns are being fixed/disclosed? This is the result of decades of security debt from poor software engineering. The labs' models are not creating net-new security risk, the risk has always been there and we're finally able to uncover and fix it.

[Mikael Brockman @meekaale]

memory safe PHP 8.4.25 built with Fil-C as Docker images (compatible with official php images) and as a single relocatable directory for Linux that doesn’t depend on any system libraries

https://t.co/d8CAfeAVTO

[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

[Bartosz Brachaczek @bbrachaczek]

@mitchellh @CadamSAFC I did it a couple days ago and was auto approved right away. Though the URL I got in Codex was wrong (it was for corporate) and I needed to google it instead.

[Mitchell Hashimoto @mitchellh]

Fork your dependencies, trim them to only your use case, never update unless it breaks for your users. I’ve been vocal about this for 10+ years. I’ve always said that updating is way riskier than latent bugs (which can be tracked and CVEs monitored).

If you are updating a dependency, it’s on you to analyze every single commit in the full transitive set of dependencies. If you dont see anything compelling, dont update!

I remember at HashiCorp once in awhile an engineer would try to update a dep or replace a DIY lib with an external one and id always ask “show me the commit we need.” Dont update for the sake of it.

Feeling pretty swell about this mentality with all the supply chain attacks happening.

[Bartosz Brachaczek @bbrachaczek]

@ivanrouzanov What about implicit conversions from void*? Most C code has no chance of compiling as C++ unless you write a lot of casts that are redundant in C.

[Bartosz Brachaczek @bbrachaczek]

@PeterAttiaMD I support you. You absolutely don't deserve these mindless accusations.

[Bartosz Brachaczek @bbrachaczek]

@cremieuxrecueil Are you tracking the country of the responders? Would be curious if people from countries with pit restrictions/bans have different scores. I don't think I've ever seen a pit bull except in images - not sure if I should attempt the test.

[Bartosz Brachaczek @bbrachaczek]

@ericparker @peterwildeford @Xbow XBOW say in their GPT-5 blog that they "didn't get much joy out of o3" with the same scaffolding.

[Bartosz Brachaczek @bbrachaczek]

@SnassyIcp @hormetic @pli_cachete No. Whether the boy in the description is the first or second one in BT-BT, it's still the same single BT-BT case. It's obvious when you consider what's the probability of having a BT-BT pair for a person having 2 children. It's 1/196, not 2/196.

[Sean Baxter @seanbax]

It's counter-intuitive, but we worry too much about very old code and not enough about new code. Old production code that's seen some stuff gets patched. If we adopt safe practices for new code, time will drive down vulnerabilities exponentially.
https://t.co/3HHmOADNrk

[Nat Friedman @natfriedman]

Ten months ago, we launched the Vesuvius Challenge to solve the ancient problem of the Herculaneum Papyri, a library of scrolls that were flash-fried by the eruption of Mount Vesuvius in 79 AD.

Today we are overjoyed to announce that our crazy project has succeeded. After 2000 years, we can finally read the scrolls:
This image was produced by @Youssef_M_Nader, @LukeFarritor, and @JuliSchillij, who have now won the Vesuvius Challenge Grand Prize of $700,000. Congratulations!!

These fifteen columns come from the very end of the first scroll we have been able to read and contain new text from the ancient world that has never been seen before. The author – probably Epicurean philosopher Philodemus – writes here about music, food, and how to enjoy life's pleasures. In the closing section, he throws shade at unnamed ideological adversaries – perhaps the stoics? – who "have nothing to say about pleasure, either in general or in particular."

This year, the Vesuvius Challenge continues. The text that we revealed so far represents just 5% of one scroll.

In 2024, our goal is to from reading a few passages of text to entire scrolls, and we're announcing a new $100,000 grand prize for the first team that is able to read at least 90% of all four scrolls that we have scanned.

The scrolls stored in Naples that remain to be read represent more than 16 megabytes of ancient text. But the villa where the scrolls were found was only partially excavated, and scholars tell us that there may be thousands more scrolls underground. Our hope is that the success of the Vesuvius Challenge catalyzes the excavation of the villa, that the main library is discovered, and that whatever we find there rewrites history and inspires all of us.

It's been a great joy to work on this strange and amazing project. Thanks to Brent Seales for laying the foundation for this work over so many years, thanks to the friends and Twitter users whose donations powered our effort, and thanks to the many contestants whose contributions have made the Vesuvius Challenge successful!

Read more in our announcement: https://t.co/rUlrdGXBMs

[Bartosz Brachaczek @bbrachaczek]

@foundmyfitness @QuirkCorey Whoa, never heard of that! And don't your YouTube smoothie recipes do that?

[Bartosz Brachaczek @bbrachaczek]

@robertwiblin An update dropped yesterday. Your poll would give a different result a week ago. I do get notified about security updates, but there's often a 1-2 week delay. Not sure about Play system updates, I guess they are opportunistic?

See https://t.co/WnXKLW4X51

[GrapheneOS @GrapheneOS]

The stock Pixel OS and Chrome use staged rollouts for updates. For Chrome, you can follow that at https://t.co/QQIl7cud8I. They start at a low % such as 5% and take days to ramp up to 10%, 25%, 50%. It can take a week or two for 100% rollout. We use our release channel system.

[Bartosz Brachaczek @bbrachaczek]

@elonmusk @jamesdouma @RadarMoron @JeffTutorials @karpathy That's a dangerous way to think about C when compiling it with GCC or Clang. Results in bugs like this: https://t.co/5OJeEWANxi

[Bartosz Brachaczek @bbrachaczek]

@lowleveldesign @timmisiak Besides checking in TEB, you can also check whether x86 ntdll is loaded. I believe it should be aliased to $ntwsym - at least that works in DbgShell, which is built directly on top of DbgEng.

[Bartosz Brachaczek @bbrachaczek]

@lowleveldesign @timmisiak Of the 3, only the effective can possibly be x86. But you have to SetEffectiveProcessorType first. Windbg does it for you, except when you inspect a full memory dump, where you need wow64exts.sw to be able to inspect x86 state of wow64 process.

[George Monbiot @GeorgeMonbiot]

1. As Germany reactivates its coal-burning power stations in response to the double whammy of Putin’s threat to gas supplies and the closure of its last nuclear reactors, please remember that the harms falsely attributed to nuclear can be correctly attributed to coal. Thread/

[vitalik.eth @VitalikButerin]

First priority is supporting Ukrainians. Second, and also really important, priority is supporting those Russians and Belarusians who refuse to let Putin's Z-ombie regime represent them and take action to oppose it.

[Dr Anton Howes @antonhowes]

A really quite horrifying story of persistent misdiagnosis of encephalitis for autism. Albeit rare, it’s worth knowing the encephalitis warning signs. https://t.co/W9qZIEY9EJ