We evaluated @OpenAI's GPT-6.2 Astra on FrontierCyber and CyScenarioBench, our offensive-cyber benchmarks.
On real systems, the model found and exploited new zero-days. One was a permission bypass in a popular mobile OS that let an app access private data despite lacking permissions. All findings are now in responsible disclosure.
On CyScenarioBench, it completed 9 of 11 multi-stage scenarios at least once, with a 63% average success rate.
We observed no successful attacks on fully hardened targets.
🚨 Rumors I’ve been hearing. First OpenAI, then Anthropic and Google.
OPENAI
OpenAI is in a “safety lockdown” on its frontier work, tightening safety measures and containment protocols to avoid another Hugging Face incident. Training continues under much stricter controls.
Most gains come from video games and extreme sandboxes where models must complete clear, difficult long-term goals. These force planning, experimentation, recovery from failure, and strategy discovery, driving intelligence and creativity.
Public research roles match this: Synthetic RL uses self-play and simulators; agent teams focus on sustained execution and multi-model coordination; the Recursive Self-Improvement team builds systems that generate hypotheses, run experiments, and automate research.
The internal model is still training but nearing completion. The jump is enormous. narrow superintelligence in specific areas and overall AGI-level capability. Its ability to independently handle difficult work sets it apart.
ANTHROPIC
Model 5 is training and is expected to solve extremely hard problems in mathematics, biology, and other sciences. Anthropic is pushing models toward original research contributions, complex experiments, and sustained reasoning.
Fable 5.5 is Model 4. Fable 6 is Model 5; Opus 6 will be based on Model 5. Internal capabilities will feed the next generation of both families.
Training environments are central. The Universes team creates realistic situations with ambiguity, interruptions, and delayed consequences. The model must track the objective, recognize failing approaches, and find alternatives.
Public roles align: scaling RL across model size, compute, and task duration, plus environments for chip design, verification, and complex visual work. Multi-agent research explores how larger model teams can handle major codebases and difficult mathematics.
Model 5 advances Claude toward owning entire research and engineering workflows. Its role in developing the next model grows more significant.
Anthropic expects RSI March 2027.
GOOGLE
Google is pushing Gemini 5 for a major leap in reasoning, multimodal understanding, and independent long-horizon work. Potentially surpassing all labs by end of year. Carbon topping Artificial Analysis signaled the speed of their frontier progress.
World models are a key advantage. Genie 4 generates interactive environments that evolve with agent actions, enabling vast practice in planning, testing approaches, and learning from outcomes.
Public roles highlight persistent context and personalization: agents that retain useful information, adapt to users, and execute complex workflows without losing track.
Gemini 5 delivers stronger intelligence across text, images, audio, and video. Combined with Google’s infrastructure and simulation work, the scale is enormous.
All three labs are also advancing robotics and world models. intelligence that understands physical environments, predicts action consequences, and operates through real machines.
Meanwhile, their models increasingly contribute to creating their successors. Each improvement accelerates the research behind the next.
The next three months are going to be insane.
[read about what’s happening with AI]
ME: Wow we’re actually legit fucked this is so bad.
[go outside]
ME: Wait those people are all acting like nothing unusual is happening. Maybe I was just in a paranoid bubble? Yeah it must be that. Phew.
[read more about what’s happening with AI]
ME: Wait no we ARE fucked. Shit. It’s all those people outside who are in a delusional bubble.
[go outside]
ME: Wait...
In a weird way, everybody is rooting for you. When you go to a job interview, the interviewer wants you to be amazing so they don’t have to look anymore. when you go on a date, the person wants to have an amazing date with you. If you make a piece of content, people want to see amazing pieces of content and they have no problem if it’s yours.
the instant you get into the mindset that people want you to do well, your entire world gets better. And you’ll be shocked to know this pattern keeps repeating in different areas of life.
Introducing HalachaBench: we gave AI programs from the makers of ChatGPT, Claude, Gemini and others real semicha exam questions from Israel's Chief Rabbinate, in Hebrew, with no books and no internet.
The best scored 80 out of 100. Many scored under 11. 🧵
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
https://t.co/JIhXN7OSh1
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
@DavidDeutschOxf Without it how do you protect the people from a looney government? Especially in Israel where parliament is controlled by the coalition?
We’re going to be in an insane world soon where attackers have access to capable models, while defenders (even ones *verified* through Daybreak / CVP) get refusals on patch analysis, bug finding, triage, tool building, verification, PoC development etc and cannot do their jobs (unless of course they also start using abliterated Chinese models, which is obviously a whole other set of risks!)
SITUATION DETECTED: Polymarket․com has been hacked. A 3rd party provider was compromised, and Polymarket has referred the matter to authorities. Users reported assets being withdrawn without permission. Anyone impacted will be made whole.
For anyone getting into biology, please consider taking airborne disease resistance more seriously.
It's a super important area of medicine that is a huge cause of death and sickness but easily gets forgotten. We're likely to get many more airborne plagues in the future due to the combination of not just extremely dense and interconnected society (including city and farm animals), but also accidents from bio research and offensive biotech (both from random people and from secret government programs that can easily get themselves exempted from safety restrictions), all enhanced by powerful AI.
Right now, this area is getting way less love than longevity research (which continues to be super important!), but it's an area where there are lots of unclaimed wins that we know exactly how to do (but are not yet doing) with present-day tech:
* early detection
* passive universal testing
* filtration/ventilation/UVC
* better vaccines (including making them open source and open science)
* cryptography to make all of the above privacy-preserving
We can have China-level pandemic resistance with Sweden-level limited intrusion into people's lives, and then go even further. We just have to actually do the work for it.