Incredibly bizarre thing happening to me: I've gotten really lazy in my prompting because the models are so robust to typos, word misuse, omitting words, etc.
But now I keep catching myself typing out messages the same way in Slack?
But now I keep catching myself typing out messages the same way in Slack?
59
4
433
16,198
What happens if these forks become good enough to compile your projects, and are 10x or 100x faster? Do they have staying power?
We're starting to see some agentic Rust forks (or alternative implementations) focused on iteration speed.
More-so than the specific projects, I'm interested in the ideas being explored (and how they could evolve upstream), so wanted to collate a few here...
24
1
169
18,088
Caring is the moat, you should feel visceral pain when things are bad or broken and stop at nothing to fix them
40
185
1,682
60,020
As a side quest, I got Astra to reduce uv's peak RSS for large solves by around 15-20% with a few small PRs. My favorite is via an arena checkpointing trick in our resolver.
15
1
92
7,029
Charlie Marsh retweeted
i know most of everyone’s focus is on the php bugs but this one is really important to patch quickly too!!
this same brotli vulnerability impacts a bunch of things, including express.js on node or bun with an app.use(express.json()) or the cURL client with the --compressed flag.
super proud of @k_tr4n and the rest of our team who helped find and disclose this!
this same brotli vulnerability impacts a bunch of things, including express.js on node or bun with an app.use(express.json()) or the cURL client with the --compressed flag.
super proud of @k_tr4n and the rest of our team who helped find and disclose this!
During this red-teaming exercise, we also found a vulnerability that allows for unauthenticated remote code execution in Brotli, a widely-deployed compression library included in many pieces of software such as Node.js, Bun, and Envoy.
The flaw could let an attacker run their own code on affected systems by sending or replying with malicious compressed data, without any need to create a user account. That is, an unpatched server that accepts compressed input could be taken over by a malicious client, and an unpatched client that accepts compressed input (such as “curl --compressed”) could be taken over by a malicious server.
This root cause issue was actually fixed in the Brotli source code in February, but it wasn’t known at the time that it gave rise to a security vulnerability and there was no more recent release of Brotli. Since Brotli is built into other software, any software which includes this library needs to take the update too, followed by the users of that software. We disclosed the vulnerability to Brotli and other downstream projects such as Node.js, which released or are currently releasing security updates, and Brotli cut new release versions and published a CVE.
More details on the patch and mitigations below:
patchtheplanet.org/brotli
0
7
61
4,730
Charlie Marsh retweeted
Day 19/ Small updates. Calls connect much faster for dots, changing reasoning in codex & work mid conversation does not invalidate cache anymore and a we've also updated the codex cli.
1,249
226
8,213
1,343,285
Charlie Marsh retweeted
During this red-teaming exercise, we also found a vulnerability that allows for unauthenticated remote code execution in Brotli, a widely-deployed compression library included in many pieces of software such as Node.js, Bun, and Envoy.
The flaw could let an attacker run their own code on affected systems by sending or replying with malicious compressed data, without any need to create a user account. That is, an unpatched server that accepts compressed input could be taken over by a malicious client, and an unpatched client that accepts compressed input (such as “curl --compressed”) could be taken over by a malicious server.
This root cause issue was actually fixed in the Brotli source code in February, but it wasn’t known at the time that it gave rise to a security vulnerability and there was no more recent release of Brotli. Since Brotli is built into other software, any software which includes this library needs to take the update too, followed by the users of that software. We disclosed the vulnerability to Brotli and other downstream projects such as Node.js, which released or are currently releasing security updates, and Brotli cut new release versions and published a CVE.
More details on the patch and mitigations below:
patchtheplanet.org/brotli
The flaw could let an attacker run their own code on affected systems by sending or replying with malicious compressed data, without any need to create a user account. That is, an unpatched server that accepts compressed input could be taken over by a malicious client, and an unpatched client that accepts compressed input (such as “curl --compressed”) could be taken over by a malicious server.
This root cause issue was actually fixed in the Brotli source code in February, but it wasn’t known at the time that it gave rise to a security vulnerability and there was no more recent release of Brotli. Since Brotli is built into other software, any software which includes this library needs to take the update too, followed by the users of that software. We disclosed the vulnerability to Brotli and other downstream projects such as Node.js, which released or are currently releasing security updates, and Brotli cut new release versions and published a CVE.
More details on the patch and mitigations below:
patchtheplanet.org/brotli
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing…
470
965
6,679
1,931,024
Charlie Marsh retweeted
The critical Brotli RCE vulnerability discovery came out of our large scale infrastructure code scanning efforts
Great research by @k_tr4n and @_sy1vi3 in assessing the widespread impact of this bug
Update your Node.js, Bun, Envoy, anything using Brotli ASAP!
Great research by @k_tr4n and @_sy1vi3 in assessing the widespread impact of this bug
Update your Node.js, Bun, Envoy, anything using Brotli ASAP!
During this red-teaming exercise, we also found a vulnerability that allows for unauthenticated remote code execution in Brotli, a widely-deployed compression library included in many pieces of software such as Node.js, Bun, and Envoy.
The flaw could let an attacker run their own code on affected systems by sending or replying with malicious compressed data, without any need to create a user account. That is, an unpatched server that accepts compressed input could be taken over by a malicious client, and an unpatched client that accepts compressed input (such as “curl --compressed”) could be taken over by a malicious server.
This root cause issue was actually fixed in the Brotli source code in February, but it wasn’t known at the time that it gave rise to a security vulnerability and there was no more recent release of Brotli. Since Brotli is built into other software, any software which includes this library needs to take the update too, followed by the users of that software. We disclosed the vulnerability to Brotli and other downstream projects such as Node.js, which released or are currently releasing security updates, and Brotli cut new release versions and published a CVE.
More details on the patch and mitigations below:
patchtheplanet.org/brotli
5
33
168
17,906
Charlie Marsh retweeted
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing…
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing…
1,103
1,181
13,199
7,125,361
This is really really cool.
With the caveat that I'm a complete novice (and the caveats in the post itself), I'm excited by the idea of using formal verification to prove these kinds of "universal properties" of existing software (like memory safety).
With the caveat that I'm a complete novice (and the caveats in the post itself), I'm excited by the idea of using formal verification to prove these kinds of "universal properties" of existing software (like memory safety).
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
11
1
183
12,585
As someone outside the field, it's a little hard for me to understand the gaps between what's proven here and the bigger vision.
E.g., how hard is it to extend this approach from RISC-V to x86? Is that ~impossible? "Within reach"?
E.g., how hard is it to extend this approach from RISC-V to x86? Is that ~impossible? "Within reach"?
3
0
8
1,940
Now, when we fork, we save the arena’s length as a "checkpoint". The active fork then appends records to the arena, and when we finish solving it (and move on to a sibling), we rewind to the checkpoint saved at the fork point.
1
0
1
1,097
Since we solve one fork at a time, we can reuse the same arena for the entire solve. We never need to clone it!
0
0
1
1,070
PSA: If you use Node or Bun, please upgrade even if you’ve enabled --use-system-ca!
That flag unions the system roots with the the bundled roots, so older versions will continue to accept the eMudhra cert even if your OS removes it.
That flag unions the system roots with the the bundled roots, so older versions will continue to accept the eMudhra cert even if your OS removes it.
10
28
511
23,436
Charlie Marsh retweeted
having Codex connect into my Samsung TV, get root access, and remove the eMudhra roots from the trust store while I’m on the other side of the country:
We're all hillclimbing on Rust rewrites but this could be another good hillclimb: how quickly a project can remediate an eMudhra-level incident
2
1
27
2,795